Researchers at DeepMind, a subsidiary of Alphabet, have introduced a state-of-the-art weather prediction program named “GraphCast,” which uses machine learning to forecast weather variables up to 10 days in advance.
Even more fascinating is that it does all this in under a minute. The recent Science journal report highlights GraphCast’s exceptional success, boasting a 90 percent accuracy rate that surpasses conventional weather prediction technologies.
GraphCast’s functionality relies on assimilating “the two most recent states of Earth’s weather,” incorporating variables from the present moment and the preceding six hours. Leveraging this data, GraphCast can accurately predict the state of the weather six hours into the future.
In a notable demonstration of real-world applicability, the AI algorithm preemptively identified the landfall of Hurricane Lee in Long Island a full ten days before it occurred, outpacing traditional weather prediction technologies then in use by meteorologists. The advantage lies in GraphCast’s ability to swiftly process information, surpassing traditional simulations that often lag due to the intricate physics and fluid dynamics involved in generating precise forecasts.
GraphCast excels in speed and demonstrates proficiency in predicting severe weather events, including tropical cyclones and extreme temperature fluctuations across regions. The report suggests that GraphCast can be re-trained periodically with recent data, allowing it to capture changing weather patterns, such as the effects of climate change and long climate oscillations.
The potential integration of GraphCast, or its underlying AI algorithm, into mainstream services is on the horizon. Google is reportedly exploring possibilities to incorporate GraphCast into its products, indicating a significant advancement in AI-powered weather forecasting with broader implications for various industries.
