GraphCast: Google's Next-Generation AI Weather Forecaster

Technology Edited by
GraphCast: Google's Next-Generation AI Weather Forecaster

GraphCast: Google's Next-Generation AI Weather Forecaster (Image: Google Blog)

Deepmind, Google”s artificial intelligence (AI) unit, has unveiled a groundbreaking AI-powered weather prediction tool. The new AI model, named GraphCast, has the potential to predict weather conditions up to 10 days in advance in less than a minute. Google has introduced this weather forecasting model in the research journal Science.

Google claims that GraphCast can make predictions more accurately and much faster than the High Resolution Forecast (HRES), used by the European Centre for Medium-Range Weather Forecasts (ECMWF). HRES is regarded as the industry”s gold-standard weather simulation system. GraphCast is also able to deliver earlier warnings of extreme weather events. “It can predict the tracks of cyclones with great accuracy further into the future,” Google mentioned in its blog post published on November 14. GraphCast also supports applications including predicting extreme temperatures and atmospheric rivers.

This AI model, based on machine learning and Graph Neural Networks (GNNs), mainly takes “the two most recent states of Earth’s weather” as input. It relies on the current state of the weather and six hours earlier to predict the next state of the weather six hours ahead. Google also informed that GraphCast is already being used by several weather agencies, including ECMWF. Most importantly, this AI weather forecaster is trained on decades of historical weather data. Google said that they adopted this process to learn a model of the cause-and-effect relationships that govern how Earth’s weather evolves, from the present into the future.

At the same time, reports suggest that the tech giant is also looking for possibilities to incorporate GraphCast into its range of products. As part of making GraphCast more accessible, Google has open-sourced the model”s code, permitting forecasters and scientists worldwide to explore the possibilities with this innovative technology.