BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260624T171342Z
LOCATION:Bldg. 6 - 002
DTSTART;TZID=Europe/Stockholm:20260701T153000
DTEND;TZID=Europe/Stockholm:20260701T160000
UID:submissions.pasc-conference.org_PASC26_sess122_msa241@linklings.com
SUMMARY:Multi-Domain: Improving Generalization Across Scales and Regions
DESCRIPTION:Sophie Buurman (KNMI), Aram Farhad Shafiq Salihi and Even Mari
 us Nordhagen (Norwegian Meteorological Institute), Mario Santa Cruz (ECMWF
 ), Michiel van Ginderachteren (RMI Belgium), and Thomas Nils Nipen (Norweg
 ian Meteorological Institute)\n\nThe domain of weather forecasting is curr
 ently undergoing a significant transformation driven by advances in machin
 e learning. Following these developments, high-resolution regional models 
 have emerged. Among these regional models is the stretched-grid model (SGM
 ), a global model with an increased spatial resolution over a region of in
 terest. Building on SGMs and generalizing the idea to incorporate more hig
 h-resolution data, while avoiding intermediate fine-tuning and transfer le
 arning steps, high resolution predictions can be achieved for any domain i
 n Europe. We propose a probabilistic multi-domain model, introducing a dyn
 amic way of training across domains and resolutions, by alternating betwee
 n different global and regional data and its corresponding graph across di
 fferent spatial regions, grid-types and resolutions. Probablistic multi-do
 main DDMs have the potential to provide a computationally cheap solution f
 or ensemble modelling at hectometric scale, motivating Task 330141 of the 
 Destination Earth Weather-Induced Extremes Digital Twin project. We utiliz
 e the multi-domain training approach based on dynamical graph training, wi
 th the aim of providing ensemble forecasts of extremes on a hectometric sc
 ale. We finetune the model on 200+ hectometric datasets of the Extremes DT
  triggered by forecast extreme events, resulting in the first on-demand au
 toregressive ensemble model of its type at hectometric scale.\n\nDomain: C
 limate, Weather, and Earth Sciences, Physics, Computational Methods and Ap
 plied Mathematics\n\nSession Chair: Oliver Fuhrer (MeteoSwiss, ETH Zurich)
 \n\n
END:VEVENT
END:VCALENDAR