The machine learning-based weather forecasting program, developed by researchers at the British company DeepMind, known as “GraphCast,” is capable of predicting weather variables for up to 10 days in less than a minute with an accuracy of up to 90%.
The AI-powered weather forecasting program benefits from understanding the “latest two states of the Earth’s weather,” which include variables that occurred during a test six hours ago. Using this information, GraphCast can predict the weather within six hours.
In terms of practical performance, artificial intelligence has demonstrated its real-world application. The tool forecasted the arrival of Hurricane “Lee” on Long Island ten days before it occurred, while traditional weather forecasting technologies used by meteorological experts at that time lagged behind.
Predictions made through standard weather simulation processes may take longer due to their traditional nature, as models must take into account complex physics and fluid dynamics to make accurate forecasts.
The weather-predicting algorithm does not only excel in forecasting weather patterns in terms of frequency and scale but GraphCast can also predict severe weather events, such as tropical hurricanes and extreme heatwaves in regions. Furthermore, the algorithm can be retrained using recent data, and scientists believe that improvements are expected mainly in predicting weather pattern fluctuations that coincide with major changes and climate variation.
In the near future, the GraphCast service may appear, or at least the foundation of the artificial intelligence algorithm supporting its forecasts in additional services. According to a Wired report, Google may be exploring ways to integrate GraphCast into its products. There are also calls to develop models capable of providing more accurate readings of the timing of severe weather occurrences and, more importantly, predicting hurricane intensity.
