Creators of the new technology say it could revolutionise forecasting.
Google Deepmind’s new technology allows 10-day forecasts to produced in less than a minute. Deepmind claims that the technology is able to produce weather forecasts for 10 days with an unprecedented level of accuracy.
Forecasts produced by the GraphCast System are not only more precise, but also more efficient. This means they can be done more quickly and using fewer resources.
It can also be used to spot extreme weather, such as cyclones. It will alert you early if there are any floods or extreme temperatures. Google says that it can save lives by helping people prepare better.
Weather forecasts are currently based on a method called Numerical Weather Prediction. This system combines physics equations and computer algorithms, which run on supercomputers. This requires extensive computing resources and detailed expertise from weather forecasters.
This new system is part of a wide range of technologies which instead use deep-learning. It does not use physical equations but instead learns from the weather data. Then it uses this to model the changes in Earth’s climate over time.
The model required extensive training, as it was based on decades worth of weather data. Now that the model is in place, it can drastically reduce the resources needed to predict the weather. 10-day forecasts only take one minute on a single machine. This process would otherwise take hundreds of computers and hours.
The system’s creators report in a new paper published in Science that it was able, in 90% of the tests, to produce more accurate predictions than the traditional gold standard system.
The system can also detect extreme weather even when it is not trained. It predicted the path for Hurricane Lee in September nine days earlier than traditional forecasts.
Deepmind stated that GraphCast’s predictions of extreme temperatures may be especially useful in light of the climate crisis. The system predicts areas where heat waves will occur above historical high temperatures. This allows people to prepare and anticipate heatwaves.
Deepmind will also open-source the system, so that others can use it. Deepmind stated that this could help develop new tools and conduct research in order to tackle environmental issues.