Artificial intelligence: Google's GraphCast wants to revolutionize predictions

the essential
Google subsidiary Deepmind has developed GraphCast, an AI-based weather forecasting tool. It is able to provide 10-day weather forecasts with unprecedented accuracy in less than a minute.

At the time Météo-France deployed Alpha last November, web giant Google was showing spectacular weather forecasting results with its GraphCast tool, developed by its subsidiary Deepmind. “In an article published in Science, we present GraphCast, a state-of-the-art artificial intelligence model capable of making medium-range weather forecasts with unprecedented accuracy. GraphCast predicts weather conditions up to 10 days in advance more accurately and much faster than the industry's gold standard weather simulation system: High Resolution Forecasting (HRES), produced by the European Center for Medium-Range Weather Forecasts (ECMWF),” Google explained.

And the company MountainView shows the prospects of its tool. “GraphCast can also provide early warnings of extreme weather events. It can predict the path of cyclones with high accuracy in the future, identify atmospheric rivers associated with flood risks, and predict the occurrence of extreme temperatures. This capability has the potential to save lives through better preparedness. GraphCast takes a major step forward in artificial intelligence for weather forecasting, delivering more accurate and efficient forecasts and paving the way to support decision-making that is central to the needs of our industries and societies.” , Google explained, hailing the fact that its GraphCast model, available as open source to scientists around the world, has already been tested by the ECMWF.

Read also:
INTERVIEW. Weather: 'Artificial intelligence perfectly predicted Hurricane Lee's track,' says meteorologist Joël Collado

In September, a publicly available version of GraphCast, developed on the ECMWF website, accurately predicted about nine days in advance that Hurricane Lee would make landfall in Nova Scotia.

40 years of data

The innovation of GraphCast is that where classic forecasters mobilize supercomputers to calculate the evolution of winds, temperatures, humidity and translate assumptions mathematically, the Google tool is based on deep learning, that is, using data rather than physical equations to creating a weather forecasting system. .

However, Google is not opposed to the two methods. “Essentially, GraphCast and traditional approaches go hand in hand: we trained GraphCast on four decades of weather analysis data, from the ECMWF ERA5 dataset. This crowd is based on historical weather observations such as satellite imagery, radar and weather stations using traditional numerical forecast times to fill in the gaps where observations are incomplete, to reconstruct a rich record of global weather history,” Google explains.

Read also:
CASE. After a series of hiccups: Météo France coughs up, forecasts take a turn for the worse

The performance of GraphCast is in any case spectacular. In a comprehensive performance evaluation against the deterministic reference system, HRES, GraphCast provided more accurate predictions in over 90% of the 1,380 test variables and shorter prediction times.

But AI also has its weaknesses: the high cost of its “training”, the management of uncertainty or its geographical accuracy (every 28 km where Météo-France can reach up to 1 km).

Leave a Reply

Your email address will not be published. Required fields are marked *