Google DeepMind launches 'GenCast,' an AI model that provides faster and more accurate weather forecasts up to 15 days ahead



Google DeepMind announced the weather forecasting AI ' GenCast ' on December 4, 2024. By using

ensemble forecasting , it is possible to predict the weather up to 15 days in advance, and it has been reported that it has demonstrated higher accuracy than the forecasting system of the European Centre for Medium-Range Weather Forecasts (EMCWF), which has been adopted by many countries and regions.

Probabilistic weather forecasting with machine learning | Nature
https://www.nature.com/articles/s41586-024-08252-9

GenCast predicts weather and the risks of extreme conditions with state-of-the-art accuracy - Google DeepMind
https://deepmind.google/discover/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/



Google's DeepMind tackles weather forecasting, with great performance - Ars Technica

https://arstechnica.com/science/2024/12/googles-deepmind-tackles-weather-forecasting-with-great-performance/

Weather shapes human decision-making, safety and the way we live, and accurate and reliable weather forecasts are crucial in recent years as climate change leads to more extreme weather events. However, weather cannot be perfectly predicted, and forecasts several days in advance can be uncertain.

So many scientists and weather agencies use a type of numerical forecasting called 'ensemble forecasting,' which projects a variety of possible weather scenarios for a given weather event, giving them a complete picture of the likely weather conditions and the likelihood of each scenario over the next few days and weeks.

GraphCast, developed by Google DeepMind, has learned from the past 40 years of weather observation data contained in the weather observation dataset ' ERA5 ' published by the European Centre for Medium-Range Weather Forecasts (ECMWF) , and can predict future weather conditions quickly and accurately.



Testing on 2019 weather data, GenCast was able to predict the weather with higher accuracy than EMCWF's ensemble forecast system. It also took just eight minutes to create a 15-day forecast on a Google Cloud TPU v5, and all forecasts in the ensemble could be created simultaneously. In contrast, traditional physics-based ensemble forecasts require supercomputers with tens of thousands of processors to run for hours to create them.

The following is a diagram showing GenCast's prediction of the path of

Typhoon Hagibis, which struck Japan in October 2019. Seven days before Typhoon Hagibis made landfall in Japan, various paths were predicted, including a route along the west side of Kyushu, but as the typhoon approached the Japanese coast five days and three days before, the range of predicted paths narrowed, making reliable and accurate forecasts possible.



Such highly accurate weather forecasts are useful for disaster prevention and safety, and also play an important role in other aspects of society, such as renewable energy planning. In fact, improving the accuracy of wind power forecasts could directly improve the reliability of wind power as a sustainable energy source and accelerate its adoption. In addition, a proof-of-principle experiment analyzing forecasts of total wind power generation by wind farms around the world reported that GenCast was more accurate than EMCWF's ensemble forecast system.

Google DeepMind said, 'We will soon release real-time and historical forecasts from GenCast and previous models, allowing anyone to integrate these results into their own AI models and research workflows. We are eager to engage the broader weather community, including academic researchers, meteorologists, data scientists, renewable energy companies, and organizations focused on food security and disaster response.'

Google DeepMind has published the source code for GenCast and

GraphCast on GitHub.

GitHub - google-deepmind/graphcast
https://github.com/google-deepmind/graphcast

Related Posts:

in Software, Posted by log1r_ut