Intelligent Earth Seminar: Ferran Alet (Google DeepMind)

8 October 14:00
Seminar Room 1a, Doctoral Training Centre, 1-4 Keble Rd
ferranalet

Ferran Alet is a Research Scientist at Google DeepMind working on generative models for weather predictions. He holds a PhD in Computer Science from CSAIL, MIT. He aims to better understand and improve machine learning generalisability by leveraging techniques from meta-learning, learning to search, program synthesis, as well as insights from the mathematical and physical sciences. 

Skillful probabilistic weather and cyclone forecasting with machine learning

Machine learning (ML)-based weather models have rapidly risen to prominence due to their greater accuracy and speed than traditional forecasts based on numerical weather prediction (NWP), recently outperforming traditional ensembles in global probabilistic weather forecasting. We will present FGN, a simple, scalable and flexible modelling approach which significantly outperforms the current state-of-the-art models. FGN generates ensembles via learned model perturbations with an ensemble of appropriately constrained models. It is trained directly to minimise the continuous rank probability score (CRPS) of per-location forecasts. It produces state-of-the-art ensemble forecasts as measured by a range of deterministic and probabilistic metrics, makes skilful ensemble tropical cyclone track predictions, and captures joint spatial structure despite being trained only on marginals. 

We will also show preliminary results from the first single model to be skilful at both cyclone track and intensity forecasting, being used by agencies such as the US National Hurricane Centre.

For more information: 

- https://arxiv.org/pdf/2506.10772

- https://deepmind.google.com/science/weather