A program called “GraphCast,” which was developed by DeepMind researchers and is based on machine learning, can predict the weather for the next 10 days in less than one minute. In the research report, the researchers emphasize that GraphCast outperformed traditional weather forecasting technologies with a 90% confidence level.
The AI-powered weather forecasting program works by taking into account “two current states of the weather on Earth”, which includes readings from the time of the test and six hours before. Based on this data, GraphCast can predict what the weather will be like in the next six hours.
In practice, artificial intelligence has already proven its use. The tool predicted Hurricane Lee’s impact on Long Island ten days before it happened, while traditional weather forecasting technologies used by meteorologists at the time lagged behind. Predictions made by standard weather simulations can take longer because the models must account for complex physical phenomena to make predictions accurate.
Not only does the weather forecasting algorithm outperform traditional weather forecasting technologies in terms of speed and scale, GraphCast can also predict severe weather events, including tropical cyclones and extreme heat waves over regions.
And because the algorithm can be retrained based on the latest data, scientists believe the tool will only get better at predicting differences in weather patterns that coincide with larger climate changes.