EARTH SYSTEM PHYSICS SEMINAR

[Pandora Malchose]

EARTH SYSTEM PHYSICS SEMINAR

Wednesday, 9 December 2009
Seminar Room, LB First Floor, 14:30 hrs

Stefano GALMARINI
Joint Research Center
Institute for Environment and Sustainability
Ispra, Italy


ENSEMBLE TECHNIQUES FOR ATMOSPHERIC DISPERSION AND AIR QUALITY MODELS: A 
CRITICAL REVIEW


The use of ensembles of model results in atmospheric transport and 
dispersion applications is becoming more and more popular. The 
properties of ensembles have been demonstrated in various situations and 
case studies. Often in the scientific literature we read the mantra: 
"/... the performance of ensemble of models is shown to be 
(systematically) superior to that of individual models/" or "... /the 
performance of ensemble of models is shown to be (systematically) 
superior to that of best performing member/". However this is a result 
of heuristic approaches and practices where no a priori investigation on 
the suitability of the constituted ensemble is presented. Given N model 
simulations of to the same case study; can we always be sure that 
averaging them will produce a better result? Can we just average the 
results from any model or should you be selective and specify the 
characteristics of these results? Can one produce N slightly different 
results with one model or would it be better to use different models? 
How do you define that a model is different form another? And what about 
the mythological model independence? Can we expect two models that try 
to describe the same processes starting from similar mathematical 
descriptions to be independent? By means of analytical considerations 
and practical examples these questions are addressed showing that there 
are exceptions to the observed behaviors and ways to get an optimal 
ensemble performance with yet smaller efforts. These results will be put 
in a general context of the research in ensemble dispersion with 
emphasis to weaknesses and strengths and to where more effort should be 
devoted.


Everyone is most welcome.