DeepMind’s AI Forecasting Surpasses Gold Standard Model

In a thorough performance evaluation compared to the industry-standard NWP system, the High-Resolution Forecast (HRES), GraphCast demonstrated superior accuracy in over 90 percent of tests.

When focusing on the Earth’s troposphere, where most noticeable weather events occur, GraphCast outperformed HRES in an impressive 99.7 percent of test variables.

The Google DeepMind team was particularly impressed by GraphCast’s ability to identify dangerous weather events without specific training. By incorporating a hurricane tracking algorithm into GraphCast, the AI program immediately improved its accuracy in identifying and predicting storm paths.

GraphCast made its public debut in September through the European Center for Medium-Range Weather Forecasts (ECMWF), organisation behind HRES.

During this debut, GraphCast successfully predicted the trajectory of Hurricane Lee nine days before it made landfall in Nova Scotia.

In comparison, existing forecast programs not only showed less accuracy but also identified Lee’s destination in Nova Scotia only six days in advance.

“Pioneering the use of AI in weather forecasting will benefit billions of people in their everyday lives,” Lam emphasised on Tuesday, highlighting GraphCast’s potential significance in the face of increasingly devastating events linked to climate collapse.

“Predicting extreme temperatures is of growing importance in our warming world,” Lam continued.

“GraphCast can identify when temperatures are expected to surpass historical highs for any location on Earth.

This is particularly useful in anticipating heatwaves, which are becoming more common and pose disruptive and dangerous threats.”

Google DeepMind’s GraphCast is already accessible through its open-source code, and ECMWF plans to continue experimenting with integrating the AI-powered system into its future forecasting efforts.