Google has unveiled GraphCast, a powerful new artificial intelligence (AI) that can make weather forecasts more accurately than existing best tools. In only one minute, the technology on a single machine made accurate predictions up to 10 days in advance, a task that usually takes many supercomputers hours to achieve.
In tests, GraphCast running on a single Google TPU v4 machine was compared to the current gold-standard for weather prediction – High Resolution Forecast (HRES), a simulation system running on supercomputers. It was able to make 10-day forecasts in under a minute, and was more accurate than HRES on 90% of the test variables and forecast lead times. When the models were focused on the troposphere – the lowest layer of the atmosphere, where accurate predictions are most useful and applicable to life – GraphCast outperformed HRES 99.7% of the time.
The new AI was trained on 40 years’ worth of weather reanalysis data, gathered by satellite images, radar and weather stations. GraphCast takes the state of the weather six hours ago and the current state, and uses its data to predict the weather state six hours from now. From this, it can project forward in six-hour increments to build a forecast up to 10 days ahead.
The research was published in the journal Science.