/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

UK Met Office and DeepMind partner to improve short-term weather forecasts using AI, particularly predictions of storms and heavy rain

Financial Times Clive Cookson

Context & Ripple Effects

This 2021 partnership sits at the start of a five-year arc in which the Met Office systematically re-platformed forecasting onto AI. Months earlier it had committed to a weather supercomputer built with Microsoft for higher-resolution physical models; teaming with DeepMind adds a machine-learning track aimed specifically at short-term storm and heavy-rain prediction, where minutes matter.

The bet paid off along a traceable line: DeepMind's GraphCast model went on to beat the best conventional systems for three-to-ten-day forecasts, and by 2026 the successor WeatherNext model was accurate enough on storm tracks and intensity to be open sourced. A 2025 Financial Times survey of the Met Office alongside Google DeepMind, Nvidia, Microsoft, IBM and startups shows the field has crowded considerably since this announcement.

First-order effects

  • The Met Office gains DeepMind's learning-based methods for its hardest operational problem — imminent storms and heavy rain — while DeepMind secures a national forecaster's observational data and an official deployment channel.

Second-order effects

  • Microsoft's position inside the Met Office via the supercomputer program means two large tech firms now compete for the same institutional customer, and rivals like Nvidia, IBM and forecasting startups named in the later coverage are pushed to court other national agencies.

Third-order effects

  • If GraphCast's accuracy lead and WeatherNext's open-source release are the template, forecasting splits into a commodity AI-model layer any agency or developer can adopt, with value migrating to proprietary data, validation and operational integration — the structure this Met Office–DeepMind pairing anticipated.
  • The deal also establishes the 'state-compatible AI lab' pattern: a frontier lab embedded with a public institution to earn legitimacy and domain data, which other governments will be pressed to replicate with their own chosen labs.

The trend: National weather agencies are moving from physics-only numerical models to hybrid pipelines where AI labs supply the predictive engine — a shift that began with partnerships like this one and ended with open-source models outperforming conventional systems.

Discussion

  • @shakir_za Shakir Mohamed on x
    We've worked for a few years on the problem of nowcasting, making short-term weather predictions ☔️. I think this is a powerful example of what is possible with generative models 🚀🤩. Excited to share our paper in @Nature today🎉🥳. https://www.nature.com/...
  • @mirowskipiotr @mirowskipiotr on x
    You can read about our carefully-constructed GAN-based generative model that uses radar observations to provides probabilistic forecasts at 1km resolution, extensively evaluated up to 90min on UK and US data (in the paper). https://www.nature.com/...
  • @amyfreeze7 Amy Freeze on x
    Better science will allow for better decisions in severe weather scenarios. https://www.nature.com/... #nature
  • @hillbig Daisuke Okanohara on x
    They use a deep generative model for high-resolution precipitation nowcasting from radar, trained with temporal/spatial discriminators and grid cell regularizers. The trained model achieves accurate, fast, and non-blurry predictions. https://www.nature.com/...
  • @deepmind @deepmind on x
    Learn more about this work: https://dpmd.ai/... Access the code behind the model: https://dpmd.ai/... 4/4
  • @financialtimes @financialtimes on x
    The UK's Meteorological Office has partnered with AI company DeepMind to focus on ‘nowcasting’, a project to pinpoint the timing, location and intensity of rain at high resolution up to two hours ahead https://www.ft.com/...