Google's DeepMind has introduced a new weather forecasting model called GraphCast, which boasts a remarkable 90% accuracy rate, outperforming traditional systems. This machine learning model relies on historical data instead of complex simulations, simplifying computations and saving energy. GraphCast starts with current weather conditions and data from six hours ago, making predictions for the next six hours and then feeding those predictions back into the model to provide longer-term forecasts. It even excels at predicting extreme weather events, such as tropical cyclones, despite not being specifically trained for them. The researchers emphasize that GraphCast is intended to complement traditional meteorological methods rather than replace them.

‍

Google DeepMind's GraphCast is a revolutionary weather forecasting model that outperforms traditional numerical weather prediction systems with over 90% accuracy. Unlike conventional models that rely on complex simulations, GraphCast uses historical data, reducing computational complexity and energy consumption. It starts with current and past weather data, makes short-term predictions, and iteratively refines them for longer-term forecasts. Notably, GraphCast excels at predicting extreme weather events, despite not being specifically designed for them.