The machine learning-based weather forecasting program called “GraphCast” developed by DeepMind researchers reports weather variables in less than a minute for 10 days. in time can guess. In a report, scientists highlight that GraphCast outperforms traditional weather forecasting technologies with a 90% accuracy rate. The AI-powered weather forecast program works by taking the “two most recent states of Earth’s weather,” which include variables at the time of testing and six hours earlier. Using this data, GraphCast can predict what the weather will be like in six hours.

In practice, AI has already demonstrated its applicability in the real world. While the tool predicted the recent Hurricane Lee’s landfall on Long Island 10 days before the event, traditional weather forecasting technologies used by meteorologists at the time were left behind. Predictions made with standard weather simulations can take longer because traditionally models have to take complex physics and fluid dynamics into account to make accurate predictions.

The weather forecasting algorithm performed well in predicting weather patterns in terms of speed and scale. Because the algorithm can be retrained with the latest data, scientists believe the tool will be better at predicting swings in weather patterns that coincide with larger changes consistent with climate change.