AI weather forecasting just took a big step forward

Axios

24-03-2025

The field of AI-driven weather modeling is advancing at a rapid pace, as illustrated by a new model that has critical advantages over other approaches.
Why it matters: Applying artificial intelligence to weather prediction holds the promise of significantly advancing forecast precision, reliability and delivery to the developing world.
It could augment the role of human weather forecasters, providing them with another tool for forecasting extreme weather events as well as routine conditions.
Driving the news: The new model is the result of an international effort among the University of Cambridge, Alan Turing Institute, Microsoft Research and European Centre for Medium-Range Weather Forecasts (ECMWF).
Zoom in: The new model, detailed in a study in the journal Nature, is known as Aardvark Weather. It offers what its creators call an "end-to-end AI forecasting system."
Previous AI models developed by technology giants like Nvidia and Google take in real-world observations and apply AI methods to predict how weather conditions would unfold over time.
These models don't require supercomputers and can be run at a fraction of the time of regular physics-based numerical models like the U.S. Global Forecast System, or GFS.
Yes, but: The AI models developed to date are still somewhat dependent on the work of traditional numerical systems at the initial step of incorporating vast amounts of weather data.
What sets Aardvark Weather apart — and may usher in a new era in AI-driven models — is that it uses a single machine-learning model that takes in observations from satellites, weather stations, ships and other sensors, and yields high-resolution global and local forecasts.
It doesn't involve traditional numerical weather models at any step of the process, setting it apart from other new AI systems.
In other words, it's a purely AI-driven weather play.
Aardvark also uses far fewer observations as inputs compared to both traditional models in use and other AI-driven ones.
For this reason and others, it may not yet be suitable for government forecast agencies.
Those agencies are generally responsible for producing forecasts with more variables, using models that assist with issuing extreme weather watches and warnings.
The intrigue: The researchers tout Aardvark's ability to result in specially-tailored forecasts while being run on a desktop computer, providing results that are available within minutes.
Importantly, they claim that even with just a fraction of the input data from current weather observations, the system outperforms the GFS model on particular variables and competes with National Weather Service forecasts made using a combination of modeling and human forecast expertise.
Perhaps the biggest breakthrough of the new model is that its simplicity and the way it's designed to learn from input data can provide a means for tailoring forecasts for specific applications and regions.
These could include forecasting wind speeds for renewable energy installations or predicting rainfall for agricultural interests.
Currently, such hyper-focused models can take many months to years to develop and require supercomputers to run.
Between the lines: The new, experimental model doesn't eliminate the need for real-world weather data gathering, conventional modeling or human forecasters.
In fact, the study underlined the importance of real-time weather data gathered from satellites to ensure forecast accuracy, for example.
It also couldn't have been developed without abundant training data that came in the form of a dataset ECMWF developed, known as the ERA5 reanalysis.
While ECMWF has been at the forefront of developing and implementing AI models, NOAA is only beginning to travel down this road in the U.S., with the American private sector moving faster to capitalize on new technologies.
The suitability of the new model to specific forecast circumstances could benefit the Global South, where high-performance computing is lacking.
What they're saying: "Aardvark reimagines current weather prediction methods offering the potential to make weather forecasts faster, cheaper, more flexible and more accurate than ever before, helping to transform weather prediction in both developed and developing countries," said Richard Turner, a study coauthor and researcher at the Alan Turing Institute and Cambridge University, in a statement.