New AI Weather Model Surpasses Leading Global Forecasting Systems

Microsoft has created an artificial intelligence (AI) model that outperforms traditional forecasting techniques in tracking air quality, weather systems, and climate-driven tropical storms, according to a study published Wednesday.

Named Aurora, the system—still not commercially available—produced 10-day forecasts and predicted hurricane paths with greater accuracy, speed, and lower cost than standard models, researchers noted in the journal Nature.

“This marks the first time an AI model has outperformed all operational centers in hurricane forecasting,” said senior author Paris Perdikaris, a mechanical engineering associate professor at the University of Pennsylvania.

Aurora, trained solely on historical data, accurately forecasted every hurricane in 2023 more effectively than agencies like the U.S. National Hurricane Center.

Conventional weather models rely on fundamental physical laws—like the conservation of mass, momentum, and energy—and demand substantial computing power.

In contrast, Aurora’s computational costs were hundreds of times lower, according to the study.

A New Era in Climate and Weather Prediction

These findings come shortly after Huawei introduced its own AI model, Pangu-Weather, in 2023. Together, these advances could signal a major shift in how global weather agencies predict everyday conditions and extreme weather events worsened by climate change.

“We’re entering a new era in atmospheric science,” Perdikaris said in a video presentation shared by Nature.

“In the next five to ten years, the ultimate goal is to develop systems that can use real-time data from satellites and weather stations to produce high-resolution forecasts anywhere on the globe,” the researchers said.

They also noted that Aurora is the first AI model to consistently surpass seven global forecasting centers in predicting the five-day path of powerful cyclones.

In one simulation, Aurora accurately predicted both the location and timing of Typhoon Doksuri’s landfall in the Philippines four days in advance—Doksuri being the costliest typhoon ever recorded in the Pacific.

Official forecasts in 2023 had predicted the storm would pass north of Taiwan.

Microsoft’s Aurora Surpasses Leading European Weather Model in Long-Range Forecast Accuracy

Microsoft’s AI model, Aurora, also outperformed the European Centre for Medium-Range Weather Forecasts (ECMWF) in 92% of 10-day global forecasts, measured at a resolution of roughly 10 square kilometers (3.86 square miles).

The ECMWF, which provides forecasts for 35 European countries, is widely regarded as the global gold standard for weather accuracy.

In December, Google revealed that its GenCast model outperformed the ECMWF’s accuracy in over 97% of the 1,320 recorded climate disasters from 2019.

These encouraging—though still experimental—results, based on real-world events, are drawing significant attention from global weather agencies.

Some agencies, including Météo-France, are now developing their own AI-based models alongside traditional physics-driven systems.

“We’re taking this very seriously,” said Florence Rabier, Director General of the ECMWF, in a statement to AFP.

She noted that their first AI-based model, released to member states in February, is about 1,000 times more computationally efficient than the standard physical model.

Although it runs at a lower resolution (30 sq km) compared to Aurora, the ECMWF’s AI model is already in operational use.

Read the original article on: Science Alert

Read more: Climate Change is Reshaping Wine Regions Worldwide, With Uneven Effects