seminario presso CNR-IMATI di Milano

[Alessandra Guglielmi <alessan@mi.imati.cnr.it>]

Il giorno 1 giugno, alle h 14.30, si terra' presso il CNR-IMATI,
sezione di Milano, Via Bassini, 15, Milano, in aula convegni (piano
terra), il seminario

    BAYESIAN METHODS FOR THE ANALYSIS OF CLIMATE OUTPUT

                          BRUNO SANSO'
			  
              University of California Santa Cruz (USA)   


Tutti gli interessati sono cordialmente invitati a partecipare. Ulteriori
informazioni sono disponibili alla pagina web http://www.mi.imati.cnr.it/.

  Cordiali saluti

  Alessandra Guglielmi


------------------------------------------------------------------
ABSTRACT.  Large deterministic models are run in order to obtain a
description of the global dynamics of climate.  There are several
statistical problems related to the analysis of the output from these
types of models.  A coarse classification is given by four types of
problems:  (a) Model validation, how much does the model output
resemble observations?  (b) Model comparison, when the model is forced
with different boundary conditions, how similar are the results?  (c)
Model calibration, what are the ranges of tuning model parameters that
produce sensible output?  (d) Data assimilation, how can data and
model output be merged into the analysis?  Statistical methods that
deal with this problems have to take into account the large sizes of
spatial and temporal domains, the large number of data points and the
fact that observations may be scarce and irregular.  Another important
issue regarding climate change is to characterize the behavior of
extremes values of atmospheric variables.
In this talk I would present examples of (a) and (c).  To illustrate
(a) I will consider the validation of a Regional Climate Model (RCM)
for Northern California.  We use process convolutions to build an
empirical statistical spatial model for temperature records of the
last 54 years.  We use predictive distributions from such model to
perform the validation.  To illustrate (c) I will consider the
calibration of three parameters used in the MIT2D model using three
different statistics that compare data to model output.
To illustrate the methods used to model extreme values I will discuss
a new approach for observations in time and space.  I will assume that
the observations follow a Generalized Extreme Value (GEV)
distribution.  The location parameters of the GEV evolve in time
through a Dynamic Linear Model (DLM).  This allows to estimate the
trend or seasonality of the data in time as well as the strength of
possible time varying associations with relevant covariates.  The
spatial element is imposed through the evolution matrix of the DLM
where a process convolution form is adopted.  Examples with rainfall
and ozone data will be presented. 
--------------------------------------------------------------------

=============================================================================
NOTA:
Le norme per utilizzare il forum SIS e le istruzioni per iscrizione
e cancellazione sono disponibili all'indirizzo
.
.   http://w3.uniroma1.it/sis/forum.asp
.
L'archivio di tutti i messaggi (aggiornato al giorno precedente)
e' disponibile all'indirizzo
.
.   http://www.stat.unipg.it/cgi-bin/wilma/sis/
.
=============================================================================