Kolloquium über anwendungsorientierte Statistik
Universität und ETH Zürich
Seminar für Statistik, ETHZ
Using non-stationary hidden Markov models to downscale
general circulation models
Prof. P. Guttorp, University of Washington
29. Oktober 1998, 16.15 -- ca. 17.30
Hauptgebäude der Universität, Hörsaal E 18
Abstract
Joint work with James P. Hughes, National Research Center for
Statistics and the Environment University of Washington.
Simulations of the Earth's climate typically does not capture small-scale
features such as precipitation very well. Yet, in order to be able to
assess the consequences of changes in greenhouse gas concentrations in the
atmosphere, it is important to be able to produce realistic precipitation
forecasts for local areas. A method for doing this has been developed over
the last several years at the University of Washington, and consists of
a hidden Markov model, where the hidden states (called weather states)
are assumed to develop according to a non-stationary Markov chain, with
transition probabilities dependent on atmospheric conditions such as
surface pressure and geopotential height. Since these variable are predicted
comparatively well by general circulation models, this is a stochastic way
of downscaling the precipitation aspect of these models.
Further information:
Christina Künzli,
Statistics Seminar of ETH Zurich