AI in weather science

This offers a significant efficiency and performance gain to most sectors – including for how we forecast the weather and its impacts.

As the UK’s national meteorological service, the Met Office has taken steps to accelerate the adoption and integration of these technologies across weather and climate science and services; developing our capabilities, people and partnerships [LINK] through a range of pilot projects, training activities, and focused engagements with organisations applying data science to support the continued delivery of extraordinary benefit and value.

Why are AI forecasts needed?

Over the last few decades, advances in technology have increased the power and sophistication of computer models that underpin weather and climate science and services. Recent progress in these areas has brought the world to the cusp of potentially game-changing breakthroughs in weather and climate modelling. AI and ML have the potential to drive new advances in weather and climate science to help make society better able to stay safe and thrive.

This is a vital, and timely, opportunity that may help tackle one of the gravest threats facing civilisation today: our vulnerability to extreme weather events in a changing climate – with science and technology having a pivotal role to play in ensuring we can better understand and manage the key hazards in order to become climate resilient through both mitigation and adaptation.

How can AI enhance forecasts?

While the Met Office has always been a data science organisation, and to some extent has always undertaken AI and ML activities, newer techniques of recent years – particularly deep learning – have shown impressive success in many domains such as computer vision and natural language processing, with these demonstrating their potential for wider application.

This is true across the whole process for weather and climate science and services, from observations (including quality control and gap filling), through simulation (including data assimilation and model simulation), analysis (including post-processing) to products and services (including risk forecasts and warnings).

Advances in AI and ML brings the potential to not only substantially enhance existing physics-based weather models by incorporating elements of AI and ML, but also to complementing their outputs with the use of AI-based weather models that work in a completely different way to physics-based models.

Physics-based and AI-based models

Physics-based weather prediction models, such as the operational Momentum Model, are run on large supercomputers and rely on solving complex equations (that represent the governing physics of fluid dynamics) to calculate how the atmospheric conditions will change over time from those currently observed.

In contrast, AI weather prediction models, such as the experimental FastNet model (developed by the Met Office and the Alan Turing Institute) offer an alternative to directly solving the governing physics of fluid dynamics, and instead learn patterns from very large quantities of historical data (e.g. observations and physics-based simulations). By capitalising on a vast amount of information, AI models seek to identify and utilise the inherent patterns in data that are not represented explicitly.

Once trained, these AI-based models offer significant advantages in the computational power needed at runtime, and their potential for improving the performance and resolution of the forecasts.

However, possible drawbacks include the lack of understanding of the physics, which could make it harder to forecast extreme weather events, which as becoming more frequent as a result of climate change. Therefore, it is possible the next generation of forecasts will include a mixture of AI-based and physics-based forecasting methods, with meteorologists continuing to play a crucial role in determining most-likely weather in the coming days and weeks.

Similarly, for this reason, our supercomputer will remain as important as ever – not only in continuing to run the physics-based forecasts themselves, but also in producing some of the training datasets on which the AI-based forecasts will depend.

The continuing role of meteorologists

Meteorologists will continue to play a vital role in validating and issuing forecasts and guidance.

The emergence of AI-based forecasts has the potential to revolutionise operational meteorology in much the same way that computers did when numerical weather predictions began to be produced by the Met Office in 1965.

Then, as now, operational meteorologists and subject matter experts will continue to be an integral and important component of the forecast and warnings process, playing a crucial role in analysing outputs, understanding uncertainties and communicating trusted weather insight and intelligence to our customers and the communities we serve.

The Met Office is embracing the use of AI across its workforce to ensure an efficient, effective and future-proof organisation as the UK’s national meteorological service. You can find out more about AI and Met Office People here.