BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//juliacon-2026//speaker//FSAGLX
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-juliacon-2026-TGHPQ9@pretalx.com
DTSTART;TZID=CET:20260812T150000
DTEND;TZID=CET:20260812T151000
DESCRIPTION:Physical effects in oceans span a vast range of both temporal a
 nd spatial scales\, making it difficult to ensure that a single numerical 
 model resolves all relevant processes within realistic computational limit
 s. To capture effects that occur outside of the resolved scales\, models n
 eed to include parameterizations that approximate the missing physics. In 
 this work\, we explore the use of online learning for parameterizations in
  a Julia based shallow water model. \n\nHere we use ShallowWaters.jl\, a s
 ingle layer ocean model\, to test the capabilities of online learning for 
 eddy backscatter parameterizations. Backscatter parameterizations represen
 t the influence of geostrophic eddies\, small-scale turbulent processes th
 at play a large role in ocean energy dynamics. In particular\, eddies faci
 litate the transfer of kinetic energy from small to large scales\, an esse
 ntial process for general ocean circulation. In the online learning framew
 ork our backscatter parameterization is given by a neural network (NN)\, c
 onverting ShallowWaters into a hybrid ocean model based on both physics an
 d machine learning. To train the NN parameterization to capture sub-grid s
 cale physics we use data assimilation techniques\, and in particular this 
 relies on the automatic differentiation (AD) tool Enzyme.jl to compute ful
 l-model gradients. This further expands ShallowWaters to be a fully differ
 entiable shallow water model. Different loss functions are implemented\, b
 oth spectral and state\, to determine which most effectively improves the 
 parameterization\, and the resulting NN parameterizations are compared to 
 an equation-discovery closure. This work expands on prior online learning 
 research and further advances hybrid approaches for gradient-based model c
 alibration in comprehensive\, differentiable\, ocean general circulation m
 odels.
DTSTAMP:20260502T092922Z
LOCATION:Room 3
SUMMARY:Online calibration of a Neural Network Parameterization in ShallowW
 aters.jl - Sarah Williamson
URL:https://pretalx.com/juliacon-2026/talk/TGHPQ9/
END:VEVENT
END:VCALENDAR