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:20260624T171341Z
LOCATION:Bldg. 6 - 002
DTSTART;TZID=Europe/Stockholm:20260701T150000
DTEND;TZID=Europe/Stockholm:20260701T153000
UID:submissions.pasc-conference.org_PASC26_sess122_msa246@linklings.com
SUMMARY:Towards Operational Data-Driven Regional Forecasting at Convection
 -Resolving Scales
DESCRIPTION:Carlos Osuna, Claire Merker, Alberto Pennino, Andreas Pauling,
  Daniele Nerini, Francesco Zanetta, Hugues de Laroussilhe, Jonas Bhend, Ka
 trin Ehlert, and Mary McGlohon (MeteoSwiss); Michele Cattaneo (Swiss data 
 science center); and Oliver Fuhrer and Radi Radev (MeteoSwiss)\n\nTranslat
 ing global data-driven weather models to regional, convection-resolving pr
 ediction remains a key scientific challenge. In Alpine environments, orogr
 aphic precipitation, convective initiation, and valley flows demand kilome
 ter-scale resolution and hourly to sub-hourly output to provide physically
  consistent, trustworthy forecasts of local extremes. MeteoSwiss, in colla
 boration with European partners, is developing a machine learning weather 
 prediction system based on an autoregressive graph neural network within t
 he Anemoi framework. A central component is Varda, an operational model tr
 ained on REA-L-CH1 — a 1 km reanalysis over the Alpine domain produced by 
 dynamical downscaling with ICON, driven by ERA5 boundary conditions and co
 nstrained by radar-based Latent Heat Nudging and a daily snow analysis. Va
 rda runs in a quasi-operational real-time setting, initialized from ECMWF 
 IFS and KENDA-CH1 analyses, demonstrating the feasibility of convection-re
 solving data-driven forecasting over complex terrain. High spatial and tem
 poral resolution are essential to capture the rapid onset and short lifecy
 cle of convective events. We present comparisons and evaluation of various
  data-driven approaches for convective scenarios, assessing their ability 
 to reproduce key structures and extremes, and outline future development p
 lans including ensemble prediction, observational integration, and further
  resolution increases.\n\nDomain: Climate, Weather, and Earth Sciences, Ph
 ysics, Computational Methods and Applied Mathematics\n\nSession Chair: Oli
 ver Fuhrer (MeteoSwiss, ETH Zurich)\n\n
END:VEVENT
END:VCALENDAR