AI may reshape weather and climate modelling
The Met Office has published a new framework describing how machine learning can be combined with physics-based models to make weather and climate predictions faster, more flexible and potentially more accurate.
The UK’s Met Office has laid out a strategic plan for integrating AI, specifically machine learning (ML), with traditional physics-based climate and weather models. The aim is to deliver what it calls an ‘optimal blend’ of AI-driven and physics-based forecasting.
To clarify what that blend might look like, the Met Office has defined five distinct approaches. One is the familiar independent physics-based model, which uses physical laws to simulate atmospheric dynamics, trusted but computationally intensive.
At the other end is an independent ML-based model that learns patterns entirely from data, offering far greater speed and scalability.
Between these extremes lie two ‘hybrid’ approaches: hybrid-integrated ML, where ML replaces or enhances parts of the physics model, and hybrid-composite ML, where ML and physics models run separately and feed into each other.
A fifth option is augmented ML, where ML is applied after the model has run to improve its output (for example, downscaling or refining ensemble forecasts).
However, this framework is more than a technical taxonomy; it provides a shared language for scientists, policymakers, and clients to understand how AI and traditional modelling can coexist.
It also helps guide future decisions, for example, allowing gradual adoption of ML in places where it makes sense, while preserving the robustness of well-understood physics methods in critical areas.
The move comes as ML-based weather and climate tools have shown increasing promise. For instance, in 2025, the Met Office published research showing a purely ML-based model achieved seasonal forecasting skill comparable to conventional physics-based methods, but with far lower computing demands.
For digital-policy watchers and climate analysts alike, this signals a shift: forecasting may become more dynamic, scalable and accessible, especially valuable in a changing climate where speed, resolution and adaptability matter as much as theoretical accuracy.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
