Source: Shutterstock
A new fully AI-driven weather prediction system now allows researchers to use just a desktop computer to produce weather forecasts that are tens of times faster and require thousands of times less computing power compared to traditional systems.
The new model, named Aardvark Weather, is the result of an international collaboration among researchers from the University of Cambridge, Alan Turing Institute, Microsoft Research, and the European Centre for Medium-Range Weather Forecasts (ECMWF).
The weather forecasting system, detailed in a research paper in the journal Nature, is based on a deep learning architecture that processes vast amounts of observational data to generate high-resolution weather predictions. The creators of Aardvark Weather claim their model is an “end-to-end AI forecasting system.”
Accurate weather forecasting is essential for industries like agriculture, transportation, and energy, as well as for public safety during extreme weather events. Traditional weather forecasting systems rely on Numerical Weather Prediction (NWP) models, which use physics-based equations to simulate atmospheric processes. This includes several steps, including collection of observational data and advancing the state of the atmosphere over time. Not only are these methods time-consuming, they can also need vast resources and can often be less accurate compared to newer AI-powered models.
However, even some of the new AI forecasting models depend on traditional NWP pipelines for data initialization. While AI has been able to replace the numerical solver that calculates how weather evolves over time, some of the other processes remain the same.
What sets Aardvark Weather apart is that it is designed to be a single-machine model with the ability to ingest data directly from observational sources, such as weather stations and satellites. This fully AI-driven approach means that weather predictions are now achievable in minutes on a simple desktop computer. The researchers empathize that Aardvark requires significantly fewer observational data inputs compared to conventional weather models and other AI-based systems currently in use.
“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 Professor Richard Turner from Cambridge’s Department of Engineering, who led the research. “Aardvark is thousands of times faster than all previous weather forecasting methods.”
According to the researchers, using just 10% of the input data, Aardvark Weather was able to outperform the United States national GFS forecasting system on many variables. They also claim the system matched the forecast accuracy of the United States Weather Service, which utilizes input from multiple weather models and human forecasters to deliver its predictions.
“These results are just the beginning of what Aardvark can achieve,” said first author Anna Allen, from Cambridge’s Department of Computer Science and Technology. “This end-to-end learning approach can be easily applied to other weather forecasting problems, for example, hurricanes, wildfires, and tornadoes. Beyond weather, its applications extend to broader Earth system forecasting, including air quality, ocean dynamics, and sea ice prediction.”
The researchers suggest that Aardvark Weather is not just about speed and accuracy, but also about access. The system could be used to predict the weather in developing countries and data-sparse regions around the world where they might not have access to state-of-the-art weather forecast equipment, expertise, or computation resources.
Source: Shutterstock
According to Dr Scott Hosking from The Alan Turing Institute, fully AI-based forecasting systems can help “everyone from policymakers and emergency planners to industries that rely on accurate weather forecasts.” By transitioning weather prediction from supercomputers to desktop computers, the system has the potential to democratize weather forecasting, making it more accessible to a wider range of users.
Aardvark could also be tailored for specific applications and regions. For example, it could be used for predicting rainfall for agricultural interests or forecasting wind speeds for renewable energy installations.
With Elon Musk’s DOGE reducing weather balloon launches, the need for innovative and efficient forecasting models has become more urgent than ever. Aardvark Weather offers a solution to bridge this gap.
While Aardvark Weather shows potential, it remains a new and experimental model. It hasn’t eliminated the need for real-world weather data gathering, at least not yet. In fact, the study underlined the importance of real-time weather data gathered from satellites to ensure forecast accuracy. The real-world data is used to train these models, and we would need to continue gathering more data.
The researchers that developed Aardvark Weather plan on expanding their team and deploying the system in the Global South, which has greater climate vulnerability and limited infrastructure to support traditional forecasting methods. Until the model becomes more reliable, we will have to rely on more traditional methods. However, AI is opening new doors. It is also allowing experts to focus on higher-level analysis and extreme weather event prediction.