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:20241120T082410Z
LOCATION:HG F 30 Audi Max
DTSTART;TZID=Europe/Stockholm:20240604T095100
DTEND;TZID=Europe/Stockholm:20240604T095200
UID:submissions.pasc-conference.org_PASC24_sess158_pos111@linklings.com
SUMMARY:P35 - Machine Learning Emulator of the Radiation Solver in the ICO
 N Climate Model
DESCRIPTION:Poster\n\nGuillaume Bertoli (ETH Zurich)\n\nThe computationall
 y demanding radiative transfer parameterization is a prime candidate for m
 achine learning (ML) emulation.\nIn this project, we develop an ML-based r
 adiative parameterization.\nA random forest (RF) is used as a baseline met
 hod, with the European Centre for Medium-Range Weather Forecasts (ECMWF) m
 odel ecRad, the operational radiation scheme in the Icosahedral Nonhydrost
 atic Weather and Climate Model (ICON), used for training.\nFor the best em
 ulator, we use a recurrent neural network architecture which closely imita
 tes the physical process it emulates.\nWe additionally normalize the short
 wave and longwave fluxes to reduce their dependence from the solar angle a
 nd surface temperature respectively.\nFinally, we train the model with an 
 additional heating rates penalty in the loss function.\nBecause ICON top h
 eight layers are artificial sponge layers, we use an idealized formula to 
 infer the radiation there.\nWe perform a one month ICON simulation with an
  ML radiation emulator and compare it to a simulation with ecRad.\nThe sim
 ulation with the ML solver remains accurate while the computation of the e
 ntire simulation is up to 3x faster.\nThe machine learning emulator does n
 ot seem to affect the stability of ICON on longer simulations.\n\nSession 
 Chair: Iva Kavcic (Met Office)
END:VEVENT
END:VCALENDAR