UK and Canada Developers Unveils New AI-Powered Aardvark Weather

Aardvark Weather, an innovative AI model developed by researchers from the UK and Canada, has the potential to revolutionise global weather forecasting by replacing traditional simulation methods with deep learning technology to enhance both cost efficiency and accuracy.

The team, comprising experts from the University of Cambridge, the Vector Institute at the University of Toronto, the Alan Turing Institute, and others, revealed their groundbreaking findings in a recent Nature report.

Unlike conventional models that rely on complex atmospheric physics equations, Aardvark Weather uses deep learning to generate global forecasts for key variables such as wind, humidity, geopotential, and temperature across multiple pressure levels.

Additionally, it produces localised forecasts for 2-meter temperature and 10-meter wind speed.

By leveraging deep learning’s ability to identify patterns in vast datasets, Aardvark Weather can significantly increase forecast resolution and frequency, enhancing both speed and precision.

Postdoctoral fellow at the University of Toronto’s Vector Institute James Requeima said:

“At the moment, there are some computationally expensive components in the forecasting pipeline. We've been able to replace many of these time-consuming parts with much lighter-weight models trained to perform the same tasks.”

He added:

“We found that once these machine learning components are chained together, the overall performance improves significantly. By fine-tuning the entire pipeline for the final task we're targeting, we can optimize each component not just for its isolated role, but for how it contributes to the outcome we care most about.”

The project, which also involved collaboration with Microsoft Research Cambridge, the ECMWF, and the British Antarctic Survey, aims to replace each step in the forecasting pipeline, transforming raw data into more accurate and timely weather predictions.

AI-Powered Aardvark Weather

Aardvark Weather harnesses raw atmospheric data, including pressure, temperature, and humidity measurements, to generate high-resolution forecasts on both global and local scales.

The system operates through three core neural components: an encoder, a processor, and a decoder.

Encoder: Transforms raw, unstructured observational data into a structured, gridded atmospheric representation.

Processor: Uses this gridded data to produce weather forecasts.

Decoder: Converts the forecasts into specific, localised predictions.

To optimise performance and enhance accuracy, each component is initially pre-trained on ERA5 reanalysis data, a high-quality historical dataset provided by the ECMWF.

It is then fine-tuned using real-world weather observations, ensuring the model can adapt to current conditions and provide precise forecasts.

Requeima stated:

“Data assimilation, in general, works like an autoregressive procedure. You start with the current atmospheric forecast, generated by large dynamical systems that estimate its present state. At time zero, you have this initial state. But data assimilation also needs to incorporate real-time measurements from remote sensors. So, you gather actual observations alongside the model’s forecast and adjust your atmosphere estimate accordingly.”

Lower Cost, Less Time-Consuming, and More Efficient

The report highlights Aardvark's remarkable efficiency, noting that it can produce a full global forecast in just one second using only four NVIDIA A100 GPUs.

This contrasts sharply with older models, like the European Centre for Medium-Range Weather Forecasts' high-resolution forecast, which can take hours to run.

This significant reduction in computing power opens up high-quality, customisable forecasting to regions and organisations that lack the resources for traditional numerical weather prediction (NWP) systems.

Furthermore, it allows for much quicker fine-tuning of the model.

Aardvark is part of a growing suite of AI-driven tools designed to enhance meteorological predictions, particularly in responding to extreme weather events.

For instance, during recent storms like Hurricanes Helene and Milton, which hit the US East Coast in October 2024, AI was critical in improving predictions of storm intensity.

Looking to the future, Requeima and the team plan to open-source Aardvark, expanding access to this advanced technology.

He concluded:

“I think it’s an important step toward democratizing weather modeling—making it more lightweight and accessible to the public. That’s our hope. It also represents a major advancement in end-to-end weather modeling, particularly through a data-driven, machine learning approach."