Microsoft develops AI model for predicting extreme weather


Microsoft on Monday introduced its latest innovation in AI modeling with “Aurora” – a breakthrough foundation model that can more accurately predict weather patterns, ultimately helping decision-makers worldwide prepare for extreme weather events.

The state-of-the-art Aurora is expected to revolutionize weather forecasting according to Microsoft, which announced the new AI model in a company blog post Monday.

Besides boasting 1.3 billion parameters and advanced architecture elements such as 3D Swin Transformers and 3D Perceiver-based encoders and decoders, what sets the model apart is that it was trained on “over a million hours of diverse weather and climate simulations,” the blog states.

Microsoft said the extensive atmospheric training enhances Aurora’s predictive capabilities for various weather scenarios allowing it to “develop a comprehensive understanding of atmospheric dynamics,” outperforming traditional numerical weather prediction systems.

The blog said researchers realized there was an urgent need for more accurate forecasting as climate change continues to fuel more severe weather patterns. It specifically mentioned Storm Ciarán, an unexpected windstorm that befell northwestern Europe in fall 2023, as its catalyst.

First causing heavy snowfall and tornados in the eastern half of the US and Canada, the ‘rare meteorological event’ then left a trail of destruction across France, the United Kingdom, Italy, Belgium, Spain, and the Czech Republic, killing 21 people without warning.

“Aurora presents a new approach to weather forecasting that could transform our ability to predict and mitigate the impacts of extreme events – including being able to anticipate the dramatic escalation of an event like Storm Ciarán,” the Microsoft research team said.

Aurora has versatility and diversity

The artificially intelligent model can forecast a wide range of atmospheric variables, including temperature, wind speed, air pollution levels, and greenhouse gas concentrations, and excels at making predictions even in data-sparse regions or during extreme weather.

Aurora’s pretraining on diverse datasets, as compared to just one dataset, has significantly improved the model’s performance output to a fraction of the computational cost of traditional weather prediction systems.

Microsoft atmospheric AI model Aurora
A flexible 3D foundation model of the atmosphere. Image by Microsoft.

By operating at a high spatial resolution of 0.1° (roughly 11 km at the equator) Aurora can capture intricate details of atmospheric processes, the company states.

“We estimate that the computational speed-up that Aurora can bring over the state-of-the-art numerical forecasting system Integrated Forecasting System (IFS) is ~5,000x,” the team said.

The model is also said to outperform specialized deep-learning models across different weather forecasting tasks and resolutions.

For example, In under a minute, Aurora produces 5-day global air pollution predictions and 10-day high-resolution weather forecasts that outperform state-of-the-art classical simulation tools and deep learning models.

Forecasting the unpredictability of climate change

Aurora’s groundbreaking capabilities not only promise more accurate weather forecasts but will also herald a new era in understanding and responding to our planet's changing climate, according to the researchers.

Microsoft said Aurora is expected to create a paradigm shift for future Earth Sciences foundation models, potentially leading to “comprehensive models that encompass the entire Earth system.”

Microsoft atmospheric AI model Aurora
Fast prediction of atmospheric chemistry and air pollution using data from the Copernicus Atmosphere Monitoring Service (CAMS). Aurora outperforms operational CAMS across many targets. Image by Microsoft.

Such models could “democratize access to accurate weather and climate information,” benefiting data-sparse regions – such as developing nations and the polar regions.

This could potentially transform sectors like agriculture, transportation, and disaster preparedness, “enabling communities to better adapt to the challenges posed by climate change,” the team concluded.

“We can envision a future where foundation models like Aurora become the backbone of operational weather and climate prediction systems, providing timely, accurate, and actionable insights to decision-makers and the public worldwide,” Microsoft said.