GraphCast:
Learning skillful medium-range global weather forecasting

Source: ArXiv


After developing an artificial intelligence (AI) model capable of playing and competing with the best Go players (AlphaGo), and after developing an AI program (AlphaFold) in the joint fields of bioinformatics and theoretical chemistry (predicting the structure of proteins from their amino acid sequences). Google DeepMind teams are now tackling weather prediction with their GraphCast prediction model.

Instead of presenting you with an article on the possibilities of the GraphCast prediction model. I'd like to show you how Google DeepMind's weather prediction model works.

Figure 1 - Severe-event prediction - how GraphCast and HRES compare.

In this scientific study, Google DeepMind's scientific teams present a new machine learning-based weather prediction (MLWP) model. MWLP is an alternative to current NWP models. It improves weather predictions by using historical data, which NWP models do not.

It's worth noting that the researchers are making GraphCast's source code freely available (GitHub repository).


Next
Next

AI and the Climate Emergency: Opportunities, Challenges, and Recommendations