A new artificial intelligence (AI) model has demonstrated the ability to predict major weather events more quickly and with greater precision than several of the most widely used global forecasting systems.

This model, named Aurora, has been trained using over 1 million hours of global atmospheric data, incorporating inputs from weather stations, satellite imagery, and radar measurements. According to Microsoft scientists, this is likely the largest dataset ever utilized for training a weather-focused AI model.

Aurora successfully predicted that Typhoon Doksuri would hit the northern Philippines four days before it actually made landfall in July 2023. During that time, official forecasts had indicated the typhoon would strike Taiwan — located several hundred miles away.

In addition, Aurora surpassed the performance of standard forecasting tools employed by agencies such as the U.S. National Hurricane Center and the Joint Typhoon Warning Center. It provided more accurate five-day storm paths and generated high-resolution forecasts up to 5,000 times faster than traditional weather models driven by supercomputers.

On a broader scale, Aurora outperformed existing systems in predicting weather conditions over a 14-day period in 91% of cases, according to the researchers. Their findings were published on May 21 in the journal Nature.

Future forecasting

The team behind Aurora hopes that this model and similar ones can contribute to a novel method of environmental prediction known as Earth system forecasting. In this approach, a single AI model could simulate weather patterns, air quality, and oceanic conditions simultaneously. This advancement may lead to quicker and more consistent weather forecasts, particularly in regions with limited access to advanced computing resources or comprehensive monitoring systems.

Related: Google builds an AI model that can predict future weather catastrophes

Sign up for the Live Science daily newsletter nowAurora falls into a category of large-scale AI systems referred to as foundation models — the same type of AI architecture that powers platforms like ChatGPT.

These foundation models are adaptable to various tasks because they're designed to learn general patterns and relationships from vast datasets, rather than being built for one specific function. As explained by the researchers, Aurora learns how to generate forecasts in seconds by analyzing weather data from sources including satellites, radar systems, weather stations, and simulated forecasts.

Once trained, the model can be adjusted for numerous scenarios with minimal additional input — a contrast to conventional forecasting models, which are typically developed for narrow purposes and often require retraining to adapt to new conditions.

The wide variety of data used to train Aurora not only improves its overall accuracy compared to traditional methods but also enhances its ability to forecast extreme weather events, the researchers noted.

The above is the detailed content of Hurricanes and sandstorms can be forecast 5,000 times faster thanks to new Microsoft AI model. For more information, please follow other related articles on the PHP Chinese website!