NVIDIA's FourCastNet 3 Transforms Weather Forecasting with AI

In July 2025, NVIDIA unveiled FourCastNet 3, an advanced AI-driven weather forecasting model that significantly enhances both the speed and accuracy of global weather predictions. This development is part of a broader trend where artificial intelligence (AI) is increasingly integrated into meteorology to provide more precise and timely forecasts, a necessity in the face of escalating climate change and the associated rise in extreme weather events.

FourCastNet 3 represents a substantial advancement in weather modeling by implementing a scalable, geometric machine learning approach to probabilistic ensemble forecasting. This model respects spherical geometry and accurately models the spatially correlated probabilistic nature of atmospheric processes, resulting in stable spectra and realistic dynamics across multiple scales. Notably, FourCastNet 3 delivers forecasting accuracy that surpasses leading conventional ensemble models and rivals the best diffusion-based methods, while producing forecasts 8 to 60 times faster than these approaches. For instance, it can generate a 15-day global weather forecast in just 64 seconds, marking a substantial improvement over traditional physics-based forecasting methods. research.nvidia.com

The model's computational efficiency is achieved through a purely convolutional neural network architecture specifically tailored for spherical geometry. Scalable and efficient large-scale training on 1,024 GPUs and more is enabled by a novel training paradigm for combined model- and data-parallelism, inspired by domain decomposition methods in classical numerical models. Additionally, FourCastNet 3 enables rapid inference on a single GPU, producing a 60-day global forecast at 0.25°, 6-hourly resolution in under 4 minutes. Its computational efficiency, medium-range probabilistic skill, spectral fidelity, and rollout stability at subseasonal timescales make it a strong candidate for improving meteorological forecasting and early warning systems through large ensemble predictions. research.nvidia.com

In parallel, the UK Met Office has been collaborating with The Alan Turing Institute to develop AI models aimed at improving the forecasting of extreme weather events. This partnership focuses on deploying machine learning technologies alongside traditional techniques to enhance the accuracy of predictions for events such as exceptional rainfall and impactful thunderstorms. The collaboration has led to the development of FastNet, an experimental global AI-based weather forecast system. FastNet utilizes a graph neural network to forecast weather patterns, allowing researchers to test its accuracy against existing numerical weather prediction methods. The Met Office is running the experimental FastNet system in near real-time, publishing daily illustrations of its output for research purposes. metoffice.gov.uk

The partnership between the Met Office and The Alan Turing Institute underscores a shared commitment to leveraging AI for public good. By integrating AI into weather forecasting, they aim to provide more accurate and timely predictions, which are crucial for disaster preparedness and various industries that rely on precise weather information. turing.ac.uk

The integration of AI into meteorology reflects a significant shift in the field, where machine learning models are increasingly used to process vast datasets more efficiently. This evolution is crucial for disaster preparedness and various industries that rely on precise weather information. Tech giants like Google DeepMind, Microsoft, and IBM, along with specialist AI weather start-ups, are heavily investing in AI forecasting. These investments aim to enhance short-term "nowcasting" and advance medium- and sub-seasonal predictions, providing more accurate, localized, and timely forecasts. ft.com

However, the generally optimistic outlook is tempered by challenges such as potential reductions in funding and staffing at key meteorological agencies, which could hinder progress. To address data gaps, private companies are deploying their own satellite constellations, ensuring the continuous flow of critical data for AI models. ft.com

The advancements in AI-driven weather forecasting models like NVIDIA's FourCastNet 3 and the collaborative efforts of the UK Met Office and The Alan Turing Institute signify a transformative period in meteorology. By harnessing the power of AI, these developments promise more accurate and timely weather predictions, enhancing our ability to prepare for and respond to extreme weather events in an era of climate change.