SQUIDS Seminar: Andrew McRae - Deep learning for weather forecasting
|Dates:||15 March 2023|
|Times:||15:00 - 16:00|
|What is it:||Seminar|
|Organiser:||Department of Mathematics|
|Who is it for:||University staff, External researchers, Current University students|
|Speaker:||Dr Andrew McRae|
Join us for this research seminar, part of the SQUIDS (Statistics, quantification of uncertainty, inverse problems and data science) seminar series.
Abstract: Weather forecasting has traditionally been performed using physics-based computer models. These simulate the underlying fluid PDEs, along with a raft of other physical processes, that govern the evolution of the atmosphere and the ocean. The accuracy of such models is limited by many factors, including the simplified representation of physical processes, the resolution of the underlying model grid, and the coverage and accuracy of weather observations, which are used when constructing the initial state.
In the past few years, there has been an explosion in the use of ML for weather forecasting. Much of this is "post-processing" -- using modern machine learning techniques to improve the output of existing models. I will talk about some of my own work, using generative adversarial networks to increase the resolution and accuracy of the output of an existing global forecast model.
An arguably more exciting strand of research is to use neural networks to replace existing weather forecast models entirely. Progress has been incredibly swift, and some recent efforts, from large AI companies, already claim to match or exceed the accuracy of the best traditional forecasting systems. I will summarise some of the recent progress in this area, and discuss the next steps in the development of data-driven forecast models.
Dr Andrew McRae
Organisation: University of Oxford
Travel and Contact Information
Frank Adams 2
Alan Turing Building