Google DeepMind has introduced GenCast, an AI weather prediction model that surpasses current systems. Published in Nature, GenCast is DeepMind's second recent AI weather model, following July's NeuralGCM, which integrated AI with physics-based methods and required less computing power but performed similarly to conventional forecasting. Unlike NeuralGCM, GenCast relies solely on AI, akin to how ChatGPT predicts text, by forecasting likely weather conditions using learned patterns from comparing predictions with actual weather data. It was trained on 40 years of data from 1979 to 2018 and outperformed the currently best Ensemble Forecast (ENS) in predicting 2019 weather 97% of the time, especially for wind and extreme events like tropical cyclones. This boosts wind power efficiency and disaster planning. Other tech giants are also leveraging AI for weather forecasting. Nvidia released FourCastNet in 2022, and in 2023, Huawei launched its Pangu-Weather model, focusing on deterministic forecasts.

In contrast, GenCast provides probabilistic forecasts, offering likelihood estimates, which aids in assessing varying weather scenarios for better planning. Although revolutionary, GenCast isn't replacing conventional meteorology. It relies on historical data, which might falter under changing climate conditions, and a dataset like ERA5 that stems from physics-based models. Many atmospheric variables require physics-based estimates since they aren't directly observable, necessitating ongoing data updates. As noted by Aaron Hill from the University of Oklahoma, and Ilan Price from DeepMind, consistent data input is critical for model accuracy over time. Future plans include testing models with direct observation data such as wind or humidity. Current AI models face challenges, like predicting upper tropospheric conditions or cyclone intensity due to limited data. GenCast's creators envision collaboration with meteorologists to enhance forecasts, emphasizing human expertise for integrating diverse data inputs and making informed judgment calls.