A new AI model, GenCast, predicts weather and extreme conditions with unprecedented accuracy, advancing the field of weather forecasting. The high-resolution model provides forecasts up to 15 days ahead, outperforming existing systems in predicting day-to-day weather and extreme events.
GenCast builds on previous models by using an ensemble approach, comprising 50 or more predictions that represent possible weather trajectories. The model is a diffusion-based generative AI model adapted to the Earth’s spherical geometry, learning global weather patterns from four decades of historical data.
The system was tested on historical data up to 2018 and showed better forecasting skill than the European Centre for Medium-Range Weather Forecasts’ (ECMWF) ENS operational ensemble forecasting system. GenCast delivered accurate forecasts on 97.2% of targets, with higher accuracy at lead times greater than 36 hours.
GenCast’s ensemble approach strikes a balance between confidence and uncertainty, providing decision-makers with a fuller picture of possible weather conditions. The model can produce one 15-day forecast in just 8 minutes using Google Cloud TPUs, compared to traditional physics-based systems that take hours on supercomputers.
The implications of GenCast are significant, enabling officials to safeguard lives, avert damage, and save money by predicting extreme heat and cold, tropical cyclones, and renewable energy output. As part of Google’s suite of next-generation AI-based weather models, GenCast will power user experiences on Google Search and Maps, improving forecasting of precipitation, wildfires, flooding, and extreme heat.
The model has been made open-source, with its code and weights released to foster wider collaboration and accelerate research in the weather community. Partnerships between AI and traditional meteorology will continue to enhance forecasts and better serve society, making GenCast a critical advance in AI-based weather prediction.
Source: https://deepmind.google/discover/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy