Improving the Physical Realism of AI Weather Forecasts: Scott Hosking at AI for Science

Messy Data, Fastnet and Resilient Forecasting: Scott Hosking at AI for Science At the AI for Science event at the Royal Society, Scott Hosking (Alan Turing Institute) discusses how “messy” environmental observations must be made AI ready to improve forecasting and resilience, set against rising costs and impacts from extreme weather and related national security risks. He introduces the UK Met Office partnership on Fastnet, an AI weather forecasting model trained on ERA5, explains its graph neural network and multimesh design, and describes work to reduce smoothing and grid artefacts while improving physical consistency without harming skill scores. He also presents a data sparse region approach that learns directly from fragmented observations for Sub-Saharan Africa and outlines an 18 month pilot with local agencies. Hosking closes with progress toward longer range climate prediction, use of open sourced foundation models, UK supercomputing, and the importance of interdisciplinary teams and project management. Scott Hosking is Mission Director for Environmental Forecasting at the Turing Institute. In 2018, he established the BAS AI Lab and formed a research team focused on Arctic sea ice forecasting. Working in collaboration with Fellows at the Turing, he developed the widely recognised IceNet programme, which delivered the first AI-powered pan-Arctic sea ice forecasting model. 00:00 Session Kickoff 00:16 Meet Scott Hosking 01:02 Messy Data Challenge 01:27 Extreme Events Rising 04:56 Climate as Security Risk 06:15 Turing Work Overview 07:53 AI for Science Pillars 09:13 FastNet Weather Forecasting 11:26 ERA5 Training Data 12:42 FastNet Architecture 14:58 RMSE Smoothing Limits 17:15 Artifacts and Fixes 22:23 Forecasting Data Sparse Regions 23:10 Learning from Raw Observations 27:13 Operationalizing FastNet 29:53 Toward Climate Prediction 32:14 Foundation Models Ecosystem 34:07 People and Culture Template 36:18 Thanks and Wrap Up