New AI model Aurora already better at weather prediction than current systems

A new artificial intelligence (AI) model named Aurora uses machine learning to outperform current weather prediction systems, researchers say, and hope it will help forecast extreme weather better.

Reliable weather forecasting is essential for mitigating natural disasters and supporting human progress. Traditional numerical models, although powerful, are extremely computationally expensive.

Here’s where AI enters the scene. Recent advances have already shown promise in improving predictive performance and efficiency, and now, a new model seems able to outperform all the others.

According to a study published by the journal of science Nature, the model named Aurora can accurately predict tropical cyclone paths, air pollution, and ocean waves, as well as global weather at the scale of towns or cities.

Forecasts by this large-scale model, which was trained on more than one million hours of diverse geophysical data, are offered in a matter of seconds. Plus, this type of work has proved to be a lot cheaper than traditional forecasting.

“With the ability to be fine-tuned for diverse applications at modest expense, Aurora represents a notable step towards democratizing accurate and efficient Earth system predictions,” says the study.

Needless to say, with climate change worsening, accurate predictions of extreme weather events are vital. Besides, in recent months, the US government has cut funding and fired staff at the National Weather Service, making it harder for the agency to get important alerts out in time.

Standard forecasting systems don’t use machine learning. They solve complex math and physics equations to simulate how conditions will likely change over time – but it can be extremely difficult.

“With the ability to be fine-tuned for diverse applications at modest expense, Aurora represents a notable step towards democratizing accurate and efficient Earth system predictions,”

says the study.

For instance, in 2023, official forecasts a few days in advance of Typhoon Doksuri got its path wrong, and when the storm hit the Philippines, dozens of people who weren’t evacuated died.

In a test scenario, though, Aurora correctly predicted Doksuri’s track from data collected four days in advance.

According to study coauthor Megan Stanley, a machine intelligence researcher based at Microsoft Research in Cambridge who spoke to Science News, outperforming official forecasts is extremely rare, so this is a major achievement that could save a lot of lives in the future.

Microsoft has been working with Aurora since June 2024, and in late 2023, Google introduced GraphCast, an AI model allegedly able to make 10-day weather predictions at “unprecedented accuracy” in under one minute.