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:20220812T074334Z
LOCATION:Sydney Room
DTSTART;TZID=Europe/Stockholm:20220627T133000
DTEND;TZID=Europe/Stockholm:20220627T140000
UID:submissions.pasc-conference.org_PASC22_sess117_msa195@linklings.com
SUMMARY:Fortran-Pytorch Bridge for Machine Learning Applications in Earth 
 System Models
DESCRIPTION:Minisymposium\n\nFortran-Pytorch Bridge for Machine Learning A
 pplications in Earth System Models\n\nArnold, Sharma\n\nIn recent years, m
 achine learning (ML) based parameterizations have become increasingly comm
 on in Earth System Models (ESM). They can be computationally faster and al
 low for the inclusion of computationally intensive parameterizations for s
 ub-grid scale physical processes. In our application, we want to learn an 
 ML based parameterization for sub-grid scale convective processes in ICON-
 NWP and deploy it in an online setting, i.e., coupled to the ESM at runtim
 e. To reduce the time spent in development and offline testing, we need to
  establish an efficient bridge between ICON-NWP, running in Fortran on mul
 tiple CPU nodes of the German Climate Computing Centre’s Levante HPC syste
 m, and neural networks trained in the Python ecosystem and potentially run
 ning on GPUs. Existing generic solutions to this bridging problem lack com
 patibility with the data structures of ICON and are not flexible enough wi
 th respect to the ML model architecture. In our contribution, we explore s
 trategies to call the ML model inference from within Fortran using dynamic
  libraries and MPI (message passing interface). Once operational, this bri
 dge enables scientists and developers to test ML models in an online setti
 ng, and can be extended to other parameterizations and ESMs.\n\nDomain: Cl
 imate, Weather and Earth Sciences, Computer Science and Applied Mathematic
 s
END:VEVENT
END:VCALENDAR