STATISTICS SEMINAR AT UNIVERSITY OF BRISTOL
DEPT. OF MATHEMATICS, ROOM SM3
JULY 20 AT 2:15 PM
http://www.stats.bris.ac.uk/Seminars
DR. PHILIPPE NAVEAU
(National Center for Atmospheric Research, Boulder, Colorado, USA):
"Dependence Structure of Random Wavelet Coefficients and Study of
Mesoscale Cloud Systems using Wavelets"
Abstract:
This talk dedicated to wavelets will be divided into two parts. An
application of wavelet to the field of atmospheric sciences (cloud study)
and a study of the dependence structure of random wavelet coefficients
will be presented.
Part 1:
To model the atmospheric characteristics of mesoscale clouds system in
the Western Pacific, a variety of simulated numerical models has been
generated and studied in the past. Because of the very large size of the
data set, the statistical analysis of such simulated outputs can be very
complex, and classical statistical tools are not always possible to
implement. To classify cloud systems into three distinct regimes, highly organized
squall-lines, less organized non-squall cloud clusters and scattered
convection, an algorithm based on the discrete wavelet transforms will be
presented. This type of approach lends itself to a number of different
applications in the processing of geophysical data. Examples will
illustrate the proposed technique.
Part 2:
Although wavelet models have been proposed to decorrelate signals, this
strategy must be reexamined when dealing with no-linear or non-Gaussian
processes. To study the departures from linearity and Gaussianity we
consider higher cumuants and higher spectra. New expressions for cumulants
of the random discrete wavelet coefficients are derived. Conditions on
the polyspectrum to give null wavelet cumulants are also obtained within
and across levels.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|