3-4 October 2002, INM, Madrid-Spain. Report by Jean Quiby, Coordinator of the SRNWP Program

First European Workshop on LAM EPS

Find the Agenda (with presentations) and the meeting Summary after it.

1. Agenda and presentations

Thursday
Chairman: Jean Quiby
8.45-9.15 Bartolomé Orfila (INM), Jean Quiby (SRNWP) Welcome - introductory remarks - aims of the workshop
9.15-9.45 Ensemble Forecasting - An Introduction Tim Palmer, ECMWF
9.45-10.15 Multisystem ensembles, the SAMEX experience Eugenia Kalnay, Universty of Maryland
coffee break
10.45-11.30 Plans for short-range ensemble experiments at the Met Office Ken Mylne, UKMO.
11.30-12.00 NCEP Short-Range Ensemble Forecasting (SREF) System: Current Status and Plan Jun Du and Steve Tracton.
12.00-12.30 The ARPA-SMR LEPS system and the COSMO-LEPS project. Stefano Tibaldi, Met Service of Emilia Romagna. part 1 and part 2
12.30-13.00 Short Range Ensemble Forecasting at Météo-France Jean Nicolau, Météo-France.
lunch
14.30-15.00 Predictability Issues in High-Resolution NWP and plans at MeteoSwiss MeteoSwiss, André Walser and Marco Arpagaus
15.00-15.30 HIRLAM Mini-Ensembles of Historical Heavy Rain Events Kai Sattler, DMI.
15.30-16.00 The uncertainty in the prediction of flash floods in the Northern Mediterranean environment; single site approach and multicatchment system approach Roberto Rudari, CIMA-Centro di Ricerca Interuniversitario in Monitoraggio Ambientale, Universitá di Genova, Universita della Basilicata
coffee break
16.30-16.50 Perturbation of Parametrized Tendencies and Surface Parameters in the Lokal-Modell Susanne Theis, University of Bonn
16.50-17.10 Plans and current work Krzysztof Nowinski, Bogumil Jakubiak, ICM, Warsaw University
17.10-17.30 Plans for short range ensemble forecast in INM José Antonio García Moya, INM.
Friday
Chairman: Tim Palmer
9.00-9.30 SVs in limited area models Martin Ehrendorfer, University of Innsbruck
9.30-10.00 Dynamics and predictability of the regional Eta model Stephane Vannitsem and C Nicolis, Belgian Met Institute
10.00-10.30 Limited-area ensembles using targeted SVs Inger-Lisa Frogner, Norwegian Met Institute, part 1 and part 2
coffee break
11.00-13.30

Discussions on a possible work programme for a coordinated European approach to short-range ensemble forecasts. Topics to be discussed:

  • methodologies for boundary perturbations
  • methodologies for initial perturbations
  • representation of model error
  • size of ensembles
  • validation techniques
  • case studies
  • next meeting
lunch
15.00-15.30Final Conclusions

2. Summary of the Final Discussion

How to approach the users?

Everybody recognizes that it is a very important point, but an equally difficult one. The general opinion was that the majority of today's forecasters are not keen using EPS information. What the majority still wants to know is whether the model gives precipitation for tomorrow and how much.

With a 40% probability for more than 5mm precipitation for tomorrow, they do not feel at ease. In such a case, they will most probably rely on the deterministic forecasts to know whether it rains or not tomorrow.

How the forecasters should learn to use the EPS information?

No clear general answer could be given to this question. This does not imply that EPS for the short-range cannot be used. It is much more the consequence that Short-range EPS is much too young it is just starting, to already have developed a strategy for its use.

Which kind of Short-range EPS?

Do we want to develop Short-range EPS for the daily weather forecasting or do we want an EPS focused on some weather types or weather events? The result of the discussion is that Short-range EPS should focus on severe weather.

The fact that two Consortia (UKMO and COSMO) have already taken this way - at least in the definition of their respective strategy - has maybe biased the discussion.

How to construct a Short-range EPS for severe weather?

I had the feeling that nobody really knows. We have today no technique to specifically tune an EPS to emphasize potentially dangerous weather situations.

The remark has been made that such a system would anyway be difficult to verify as severe weather is - fortunately! - a rare event.

Focus on severe weather

There are two types of severe weather:

damages on the synoptic scale:

Examples: the October 1987 storm in South-England, the Danish storm (3rd December 1999), the 26 and 27 December 1999 storms in France, Switzerland and Germany, the August 2002 floods in Germany, Czech Republic and Austria.

the flash floods caused by severe convection:

Examples: Vaison-la-Romaine 1992 (F), Brig 1993 (CH), Versilia 1996 (I)

For the flash floods, we do not have today the necessary model resolutions.

It has clearly been said that in order to have good probabilistic forecasts of strong convection, we must go to high model resolution. We shall attain the necessary resolution for an operational EPS presumably not before 10 years.

What about the stochastic physics?

The use of a stochastic physics is preferable to the multi-model technique. But in order to get in average the best results, this method requires an intensive development phase: there are so many possibilities! Nevertheless this method has been rated as very adequate for the EPS technique.

Two words on the multi-model technique

As it produces means, the multi-model technique is thought to be too conservative; it is therefore unsuitable for an emphasis on severe weather.

An EPS for severe weather should have biased pdf's

In any model, each grid point has climatological pdf's for its different variables.

A good model should obviously have for a given grid-point the same pdf's as at its corresponding geographical location (if we forget the effects of the non-resolved features).

But pdf's of EPS trimmed to enhance severe weather must be different. Do we want this? If not, it is not clear how an EPS for severe weather should be tuned.

Computation of the SV's

Today's singular vectors are computed only for a dry atmosphere. This should be changed by computing the singular vectors of an integrated system consisting of the moist atmosphere and the soil model. But this would raise the following question: how do we perturb the soil variables?