Working Group 3

 

Quantitative Precipitation Forecast II:

Quality of Precipitation Forecast, Verification and Interpretation Issues

 

Participants:

Chair: Hans Volkert

Felix Ament, Ulrich Damrath, Klaus Dengler, Barbara Fay, Joe Klemp, Nicole van Lipzig, Chiara Marsigli, Marco Milan, Antonella Morgillo, Akira Noda, Antonio Parodi, Evelyne Richard, Takehiko Satumura, Hans Schipper, Hiromu Seko, Friedrich Theunert, Stefanie Wassermann, Heini Wernli, Tanja Winterrath, Matthias Zimmer

Report: Jean Quiby

 

In Group 2, the discussion was very general and has touched numerous points. Mentioned here are only the points, which have trigged a substantial discussion.

 

Verification of model precipitation forecasts

With the point to point verifications methods (i.e. with the scores relating a grid point to an observing station), it is a well known fact that the higher the resolution of a model, the higher is the double penalty. The explanation is simple: high resolution models show more details, more isolated and distinct features, i.e. more possibilities to be wrong.

This fact renders the use of verification methods that are not affected by double penalty an absolute necessity. And such methods already exist:

-         Object oriented methods, as the SAL method (SAL = Structure, Amplitude, Location)

-         Fuzzy techniques, as neighbouring

-         Image matching methods.

 

What is a good forecast?

The group arrived at the conclusion that – formulated in this way - it is not possible to answer that question, as the quality of a forecast can only be judged through the use that is made from this forecast. The same forecast could perfectly satisfy a user and be totally insufficient for another user working in a different field.

 

Precipitation forecasts:

“Low” resolution (say 7-10 km) versus “high” resolution (say 2-3 km)

Verified by the traditional point to point methods at the respective scales of each model resolution, the high resolution forecasts will be worse.

When comparing precipitation forecasts between models with different resolutions, it is only meaningful to do it by upscaling the fields of the higher resolution models to the coarsest model resolution.

When this is done, it can be claimed – based on the experience accumulated until today – that the forecasts of wind and temperature computed at high resolution are better than the forecasts computed at the “upscaled” resolution (when the verification is made on this latter resolution).

Is this also the case for the precipitations? No participant could claim that the above statement would also be valid for the precipitations.

Note that there is another important reason to upscale high resolution model fields for verification: there are simply not enough observations at their resolution to carry out a reliable verification.

 

Why to develop high resolution, km-scale resolution models?

As it has not yet been reliably assessed that high resolution precipitations are of better quality than precipitations from coarser models, why do we continue to develop high resolution models?

Or, more directly: where are the strengths of the high resolution, km-scale models?

1.                  As already said, the temperature and, first of all, the wind forecasts are better than at coarser resolutions.

2.                  High resolution precipitation patterns show very realistic structures, very often amazingly similar to the corresponding radar patterns, although the location and the timing are not always correct. This proves that the high resolution has the capability to very well reproduce the dynamical and thermodynamical atmospheric processes.

3.                  There today in the high resolution, km-scale resolution more potential for improvement that at the mesoscale resolution (say 7-10 km). One of the reasons is explicitness of the convection, which is not (yet?) at the mesoscale properly parameterized. The absence of parameterized convection at high resolution allows for example a better use of the radar information in data assimilation.

 

Use of the probabilistic precipitation forecasts

All the participants in the discussion agreed that the acceptance of the probabilistic forecasts by the forecasters as well as by the professional users is progressing very, very slowly.

It seems that the greatest acceptance among the professional users is to be found by the hydrologists. It has been reported that in Italy (more precisely in the Regione Emilia-Romagna), ensemble forecasts for the ensemble hydrological forecasts are derived from the meteorological EPS.

 

 

For the minutes:

Jean Quiby