BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:GenCast: Diffusion-based ensemble forecasting for medium-range wea
 ther (or: How to ruin a numerical weather forecaster’s Christmas) - Andr
 ew McDonald and Kenza Tazi\, University of Cambridge
DTSTART:20240214T110000Z
DTEND:20240214T123000Z
UID:TALK212299@talks.cam.ac.uk
CONTACT:Isaac Reid
DESCRIPTION:As Santa’s elves wrapped the world’s Christmas presents in
  the final days of 2023\, a group of scientists at Google DeepMind kept bu
 sy wrapping their own gift to the numerical weather prediction community. 
 GenCast\, a diffusion model trained to produce probabilistic global medium
 -range weather forecasts at 12-hourly\, 1-degree resolution\, was publishe
 d to arXiv on 25 December 2023 and represents the latest step forward in a
  field moving at hurricane pace. Traditionally relying upon physics-based 
 models which solve systems of differential equations describing known atmo
 spheric behaviours and tendencies in discrete cells\, weather prediction h
 as seen an influx of attention from the machine learning community in the 
 past two years\, from Google DeepMind’s GraphCast to NVIDIA’s FourCast
 Net to Huawei’s PanguWeather to Microsoft’s ClimaX. At its core\, weat
 her prediction is one of image-to-image translation\, or perhaps video fra
 me prediction—where red\, green\, and blue channels of an input and targ
 et image are swapped with physical variables such as temperature\, pressur
 e\, and wind speed in an observed state and a target state some hours ahea
 d—and otherwise arbitrary pixel space represents a fixed discretization 
 of the globe in latitude and longitude. Thus\, it seems natural for the re
 sounding success of diffusion models in conditional image generation to tr
 anslate to this task. In contrast to previous approaches using ML for weat
 her\, GenCast leverages the stochastic nature of diffusion models to produ
 ce ensemble forecasts\, each generated through conditional denoising of in
 dependent noise samples\, thereby maintaining physical consistency and avo
 iding the unrealistic spatial smoothing that comes with the use of MSE as 
 a training strategy for deterministic forecasting models. In this talk\, w
 e will present a refresher on numerical weather prediction and diffusion m
 odels\, dive into the clever details of GenCast and the engineering secret
 s to its success\, compare GenCast with other contemporary work in the ML 
 for weather space\, and conclude by highlighting the many opportunities th
 at remain for ML across the geosciences.\n\nSuggested Reading: GenCast: Di
 ffusion-based ensemble forecasting for medium-range weather (https://arxiv
 .org/abs/2312.15796)
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR