UNIVERSITY OF ST ANDREWS
Statistics Seminars
_________________________________________________________________________________________________
FRIDAY 17 OCTOBER at 4 p.m. in Lecture Theatre C of the Mathematical Institute
Dr Ken NEWMAN (University of St Andrews)
"The no observation error problem in state-space models for
animal population dynamics"
FRIDAY 31 OCTOBER at 2 p.m. in Lecture Theatre D of the Mathematical Institute
Dr Sergio PEZZULLI (Department of Meteorology, University of Reading)
"Bayesian probabilistic calibration of ensemble weather forecasts"
_________________________________________________________________________________________________
The seminar on October 31 is a joint seminar with the Applied
Mathematics Division.
Tea will be available from 3.45 p.m. on October 17, and after the
seminar on October 31.
Visitors will be very welcome.
Further information from:
Dr I B J Goudie email: [log in to unmask]
_________________________________________________________________________________________________
SEMINAR ABSTRACTS
Dr Ken NEWMAN (University of St Andrews)
"The no observation error problem in state-space models for animal
population dynamics"
State-space models are a convenient framework for simultaneously
describing how population abundances and associated measurements
evolve over time. Sequential filtering procedures, such as sequential
importance sampling, are based on generating the unseen state vector
and the efficiency of such procedures depends upon how likely the
simulated state vector is given the observation. In the case of zero
observation error, the likelihood is degenerate, equalling one only
when the simulated state vector yields the observed values exactly. A
method is proposed for generating state vectors that satisfy the
constraint created by the observation. Formulations for a removal
experiment and a mark-recapture study are presented.
Dr Sergio PEZZULLI (Department of Meteorology, University of Reading)
"Bayesian probabilistic calibration of ensemble weather forecasts"
The problem of calibrating an instrumental device has a long history
in statistics and is strongly relevant in climate predictions. After
a review of statistical and metereological literature, a simple
Bayesian calibration method will be shown at work for ENSO ensemble
forecasts. The El Nino-Southern Oscillation (ENSO) is an important
large-scale ocean-atmosphere coupled phenomenon, which affects the
climate of many regions around the world. Ensemble forecasts are
batches of predictions obtained by perturbing the initial conditions
of the coupled ocean-atmosphere
dynamical model. The Bayesian approach shows sensible improvements in
the representation of posterior uncertainty. Finally,
work-in-progress will be shown for an end-user application of
ensemble forecasts, concerning electricity demand in UK.
_________________________________________________________________________________________________
|