Google DeepMind has introduced GenCast, an AI weather model outperforming the current leading systems in accuracy, as published in Nature. Unlike its predecessor NeuralGCM, which combined AI with physics-based methods, GenCast relies entirely on AI to predict weather patterns, producing probabilistic forecasts such as the likelihood of rainfall or temperature changes. Trained on 40 years of weather data, GenCast surpassed the accuracy of the Ensemble Forecast System (ENS) 97% of the time, particularly excelling in predicting wind conditions and extreme weather, crucial for wind power optimisation and disaster planning. While AI models like GenCast represent a significant advancement, they are limited by the historical data they train on and struggle with challenges such as predicting cyclone intensity. Experts highlight the continued need for human meteorologists to provide critical judgment and integrate diverse data sources. DeepMind aims to refine AI models further, exploring predictions based solely on real-time observational data.
