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:20230831T095745Z
LOCATION:Davos
DTSTART;TZID=Europe/Stockholm:20230626T112000
DTEND;TZID=Europe/Stockholm:20230626T115000
UID:submissions.pasc-conference.org_PASC23_sess104_pos133@linklings.com
SUMMARY:P12 - Building a Physics-Constrained, Fast and Stable Machine Lear
 ning-Based Radiation Emulator
DESCRIPTION:Poster\n\nGuillaume Bertoli and Sebastian Schemm (ETH Zurich) 
 and Firat Ozdemir, Fernando Perez Cruz, and Eniko Szekely (Swiss data scie
 nce center)\n\nModeling the transfer of radiation through the atmosphere i
 s a key component of weather and climate models. The operational radiation
  scheme in the Icosahedral Nonhydrostatic Weather and Climate Model (ICON)
  is ecRad. The radiation scheme ecRad is accurate but computationally expe
 nsive. It is operationally run in ICON on a grid coarser than the dynamica
 l grid and the time step interval between two calls is large. This is know
 n to reduce the quality of the climate prediction. A possible approach to 
 accelerate the computation of the radiation fluxes is to use machine learn
 ing methods. In this work, we study random forest and neural network emula
 tions of ecRad. Concerning the neural network, we compare loss functions w
 ith an additional energy penalty term and we observe that modifying the lo
 ss function is essential to predict accurately the heating rates. The rand
 om forest emulator, which is significantly faster to train than the neural
  network is used as a reference model that the neural network must outperf
 orm. The random forest emulator can become extremely accurate but the memo
 ry requirement quickly become prohibitive. Various numerical experiments a
 re performed to illustrate the property of the machine learning emulators.
 \n\nSession Chair: Elaine M. Raybourn (Sandia National Laboratories)
END:VEVENT
END:VCALENDAR