Article Details

Google DeepMind researchers have developed a machine-learning–based weather prediction algorithm called GraphCast, marking a major breakthrough in modern weather forecasting.

In less than one minute, GraphCast can predict weather variables for the next 10 days, outperforming traditional weather prediction models with an accuracy rate of up to 90 percent.

GraphCast works by analyzing two of Earth’s most recent weather states—one at the time of prediction and another from six hours earlier. Using this data, the AI model can forecast weather conditions six hours ahead, offering a major advantage in both speed and accuracy.

According to a report from Engadget on Wednesday (Nov 15, 2023), GraphCast has already proven its capabilities in real-world scenarios, such as accurately predicting the landfall of Hurricane Lee in Long Island 10 days before the event.

The algorithm’s rapid predictions are possible because GraphCast doesn’t rely on solving complex physics or fluid dynamics equations like traditional weather models.

Advanced Capabilities Beyond Speed

GraphCast’s strengths go far beyond fast predictions. The AI system can also forecast severe weather events, including tropical cyclones and extreme temperature waves.

Since it can be retrained using updated data, GraphCast is expected to improve over time—especially in predicting climate-related weather pattern oscillations.

Looking ahead, GraphCast or its underlying algorithms may become widely accessible. Google is reportedly exploring ways to integrate GraphCast into its products.

Meanwhile, the demand for better storm modeling continues to drive innovation. For example, the U.S. National Oceanic and Atmospheric Administration (NOAA) is developing a space-based supercomputer to deliver more accurate weather forecasts and improved storm intensity predictions.