Forecast in a flash: Cambridge AI researchers assess weather in 1 second

AI weather modeling takes a quantum leap forward as Cambridge researchers demonstrate a system that can match traditional forecasting accuracy in just one second on a desktop computer, compared to hours or days on supercomputers. This breakthrough, named Aardvark Weather, represents a significant shift in the weather prediction landscape by fully replacing both the computationally intensive initialization and forecasting stages that have defined meteorological science since the 1950s.

The big picture: A new AI system can produce weather forecasts in a single second on a desktop computer that rival the accuracy of traditional numerical weather prediction (NWP) models requiring massive supercomputers and hours or days of computation time.

• Developed by University of Cambridge researchers led by Richard Turner, Aardvark Weather is the first AI model to replace both the initialization and forecasting stages of traditional weather prediction.
• While previous AI weather tools from Google and DeepMind have replaced portions of traditional forecasting, Aardvark’s comprehensive approach marks a more complete paradigm shift in meteorological modeling.

Key technical details: Aardvark Weather achieves its efficiency through a radical redesign of the weather forecasting pipeline.

• The system uses just 10% of the input data that existing systems require while achieving comparable results to the latest NWP forecasts.
• Traditional initialization—the process of collating, cleaning and organizing data from satellites, balloons and weather stations—consumes approximately half of traditional forecasting’s computational resources, which Aardvark eliminates.

The limitations: Critics point to Aardvark’s relatively coarse resolution as a significant drawback for practical implementation.

• Aardvark uses a grid model with cells 1.5 degrees square, compared to the European Centre for Medium-Range Weather Forecasts’ ERA5 model with cells as small as 0.3 degrees.
• University of Manchester’s David Schultz warns this coarser resolution means Aardvark “cannot represent the extremes at all” and might miss complex weather patterns that could “blow up your forecast.”

Behind the research: The development highlights the symbiotic relationship between traditional physics-based models and newer AI approaches.

• Turner acknowledges that AI models like Aardvark still depend on physics-based systems for training data, noting attempts to train purely on observational data “didn’t work.”
• The researchers envision a future where scientists develop increasingly accurate physics-based models that then train AI systems to replicate their output more efficiently.

Looking ahead: Some researchers believe AI weather forecasting could eventually surpass traditional methods entirely.

• Oxford University’s Nikita Gourianov suggests AI may eventually create forecasts superior to NWP using only observational and historical weather data.
• According to Gourianov, success will depend on both scale and “cleverness” in data handling and neural network structure.