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:20240603T185800
DTEND;TZID=Europe/Stockholm:20240603T185900
UID:submissions.pasc-conference.org_PASC24_sess156_pos157@linklings.com
SUMMARY:P09 - Contribution of Latent Variables to Emulate the Physics of t
 he IPSL Model
DESCRIPTION:Poster\n\nSégolène Crossouard (Laboratoire des Sciences du Cli
 mat et de l’Environnement, CEA); Masa Kageyama and Mathieu Vrac (Laboratoi
 re des Sciences du Climat et de l’Environnement, CNRS); Thomas Dubos (Labo
 ratoire de Météorologie Dynamique, École Polytechnique); and Soulivanh Tha
 o and Yann Meurdesoif (Laboratoire des Sciences du Climat et de l’Environn
 ement, CEA)\n\nAtmospheric general circulation models include two main dis
 tinct components: the dynamical one solves the Navier-Stokes equations to 
 provide a mathematical representation of atmospheric movements while the p
 hysical one includes parameterizations representing small-scale phenomena 
 such as turbulence and convection. However, computational demands of the p
 arameterizations limit the numerical efficiency of the models. Machine lea
 rning offers the possibility of developing emulators, as efficient alterna
 tives to traditional parameterizations. We have developed two offline emul
 ators of the physical parameterizations of the IPSL climate model, in an i
 dealized aquaplanet configuration, to reproduce profiles of tendencies of 
 the key variables - zonal wind, meridional wind, temperature, humidity and
  water tracers - for each atmospheric column. Initial emulators, based on 
 a dense neural network or a convolutional neural network, show good mean p
 erformance but struggle with variability. A study of physical processes ha
 s revealed that turbulence was at the root of the problem. Knowing how tur
 bulence is parameterized in the model, we show that incorporating physical
  knowledge through latent variables as predictors into the learning proces
 s, leading to a significant improvement of the variability. Future plans i
 nvolve an online physics emulator, coupled with the atmospheric model to p
 rovide a better assessment of the learning process.\n\nSession Chair: Erik
  W. Draeger (Lawrence Livermore National Laboratory)
END:VEVENT
END:VCALENDAR