8/5/26 Bi Weekly Meeting at 10 AM CT
11 views
Skip to first unread message
Biplov Bhandari
Aug 4, 2026, 3:01:31 PM (10 days ago) Aug 4
to Geo-AI Working Group
Hey Geo-AI WG
Join us on 8/5/26 for a talk from Samier Merchant on WeatherNext 2
Samier Merchant is a Senior Staff Software Engineer at Google and the engineering lead for Google's weather research initiative. His work spans developing ML weather models and building the end-to-end systems that take them from research prototypes to production. He has led the development and deployment of both MetNet (nowcasting) and WeatherNext (medium-range global forecasting).
In this talk, Samier will cover WeatherNext 2, Google's most advanced AI weather forecasting system powered by Functional Generative Networks (FGN). Learn how FGN achieves state-of-the-art probabilistic skill (CRPS) -- outperforming GenCast on 99.9% of targets -- while generating ensemble forecasts 8× faster than diffusion (WeatherNext 1 Gen). He'll also discuss how WeatherNext 2 runs operationally 4× daily, producing 50+ ensemble members, and how forecasts are freely available through Google Earth Engine and BigQuery for the geospatial community.
📜 FGN Paper (core of this talk): https://arxiv.org/abs/2506.10772
📜 GraphCast: https://www.science.org/doi/10.1126/science.adi2336
📜 GenCast: https://www.nature.com/articles/s41586-024-08252-9
Excited for the talk!
Best,
Biplov
Join us on 8/5/26 for a talk from Samier Merchant on WeatherNext 2
Samier Merchant is a Senior Staff Software Engineer at Google and the engineering lead for Google's weather research initiative. His work spans developing ML weather models and building the end-to-end systems that take them from research prototypes to production. He has led the development and deployment of both MetNet (nowcasting) and WeatherNext (medium-range global forecasting).
In this talk, Samier will cover WeatherNext 2, Google's most advanced AI weather forecasting system powered by Functional Generative Networks (FGN). Learn how FGN achieves state-of-the-art probabilistic skill (CRPS) -- outperforming GenCast on 99.9% of targets -- while generating ensemble forecasts 8× faster than diffusion (WeatherNext 1 Gen). He'll also discuss how WeatherNext 2 runs operationally 4× daily, producing 50+ ensemble members, and how forecasts are freely available through Google Earth Engine and BigQuery for the geospatial community.
📜 FGN Paper (core of this talk): https://arxiv.org/abs/2506.10772
📜 GraphCast: https://www.science.org/doi/10.1126/science.adi2336
📜 GenCast: https://www.nature.com/articles/s41586-024-08252-9
Excited for the talk!
Best,
Biplov
Reply all
Reply to author
Forward
0 new messages