BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Machine learning seminar series - Probabilistic downscaling to det
ect regional present and future climate hazards
DTSTART:20200428T103000Z
DTEND:20200428T160000Z
DTSTAMP:20240523T154300Z
UID:indico-event-189@events.ecmwf.int
CONTACT:events@ecmwf.int
DESCRIPTION:\n\nHost\n\nPeter Dueben (ECMWF)\n\nSpeakers\n\nSherman Lo hol
ds a PhD in statistics and is currently working\, at the University of War
wick\, on precipitation forecasting of the UK using climate models. He has
obtained degrees in Physics and Computational Statistics & Machine Learni
ng at University College London. He is interested in applied and computati
onal statistics and has previously worked on x-ray imaging for 3D printing
\, fusion energy and chromosome imaging.\n\nRitabrata Dutta is an Assistan
t Professor in the Department of Statistics\, University of Warwick\, broa
dly working on the applications of statistical methodology in the domain o
f meteorology\, population genetics and complex bio-medical processes. He
has developed methods for Approximate Bayesian computation (ABC)\, collabo
rated with Swiss National Supercomputing Centre (CSCS) to develop a Python
package for ABC utilizing HPC infrastructure optimally and applied ABC me
thodology to learn expensive mechanistic models explaining epidemic proces
ses on network\, molecular dynamics of water\, arterial flow of blood\, vo
lcanic eruption and passenger dynamics in airports. Presently Dr Dutta is
supervising 6 PhD students and 1 Postdoc\, working broadly on methodologic
al developments of statistics and machine learning\, applications in the d
omain sciences. He is currently PI of a project on Quantifying effects of
climate change on extreme weather events via distributional downscaling fu
nded by the Alan Turing Institute (ATI) as well as a special project of EC
MWF on Data-driven calibration of stochastic parametrization of IFS using
ABC.\n\nAbstract\n\nWe present a statistical methodology to do forecasting
of precipitation at 0.1° resolution using model fields from computer sim
ulations (air temperature\, geopotential\, specific humidity\, total colum
n water vapour\, wind velocity) at a lower resolution (5/9° and 5/6° lon
/lat). Observed precipitation at the 0.1° resolution for the past 4 decad
es was used to fit our model onto it in order to do forecasting with quant
ifiable uncertainty.\n\nOur model is the compound Poisson distribution (Re
vfeim\, K.J.A.\, 1984\, Dunn\, P.K.\, 2004) which can model both occurrenc
e and quantity of precipitation as a random variable. We impose time autoc
orrelation by introducing auto-regressive and moving average terms into th
e model. Spatial dependencies were introduced by putting a Gaussian proces
s prior on the parameters.\n\nThe model fields were interpolated\, with un
certainty bars\, from low resolution to high resolution using a Gaussian p
rocess. The entire Bayesian inference (or model fitting) was done using a
Gibbs sampling scheme consisting of Metropolis-Hastings\, slice sampling a
nd elliptical slice sampling.\n\nhttps://events.ecmwf.int/event/189/
IMAGE;VALUE=URI:https://events.ecmwf.int/event/189/logo-528950275.png
LOCATION:11:30 BST
URL:https://events.ecmwf.int/event/189/
END:VEVENT
END:VCALENDAR