Google's AI Provides Better Weather Forecasts Than Current Models
Google's innovative AI technology is revolutionizing weather forecasting by providing more accurate and timely predictions than traditional models. Leveraging vast amounts of data and advanced machine learning algorithms, Google's AI analyzes atmospheric conditions, historical weather patterns, and real-time data to deliver localized forecasts with improved precision. This breakthrough not only enhances our ability to prepare for weather-related events but also contributes to better decision-making in various sectors, including agriculture, transportation, and emergency management. As climate change continues to impact weather unpredictability, Google's AI-driven approach offers a promising solution for managing risks and ensuring safety.
kimoyandex 
Google's AI Provides Better Weather Forecasts Than Current Models
Will Google's AI provide better weather forecasts than current models? DeepMind, Google's AI research subsidiary, claims that its new AI model, GenCast, is significantly more performant in predicting the weather up to 15 days in advance.
What's the Weather Like? This question, which has puzzled humanity for centuries, is answered with varying reliability by current weather models. With GenCast, DeepMind introduces a high-resolution (0.25°) model that can provide "better forecasts, both for daily weather and extreme events, than the main operational system, the ECMWF (European Centre for Medium-Range Weather Forecasts), up to 15 days in advance." This is no small feat.
More Accurate and Less Resource-Intensive GenCast is a diffusion model—a type of AI model commonly used in generating images, videos, and music—that stands out from the crowd. It has been adapted to the Earth's spherical geometry and learns to generate complex probability distributions of various future weather scenarios based on the current weather state.
The model was trained with four decades of weather data from the ECMWF archives. To evaluate GenCast's performance, DeepMind tested forecasts from 2019 based on data trained up to 2018. The result: GenCast showed better forecasting capabilities than the European Centre, with higher accuracy in 97.2% of cases and in 99.8% of cases over a horizon greater than 36 hours.
"More accurate forecasts of extreme weather conditions can help authorities protect more lives, avoid damage, and save money. In our tests on GenCast's ability to predict extreme heatwaves, cold snaps, and high wind speeds, GenCast consistently outperformed the ECMWF," Google stated.
A Significant Advantage An additional benefit is that generating 15 days of predictions with GenCast requires only 8 minutes of computation on a single Google Cloud TPU v5 (a specialized processing unit designed by Google to accelerate AI-related calculations). Traditional forecasts, like those from the ECMWF, require hours of computation on a supercomputer.
No Replacement for Traditional Models However, it is not a matter of discarding traditional weather forecasting models. First, because the data produced by these models feed the AI. Additionally, Google, which continues to work with weather agencies, sees cooperation as essential for improving forecasts and "better-serving society."
Open Model GenCast is an open model that developers and forecasters can exploit as they see fit. They can integrate it into their own models and workflows.
