By: Elizabeth Kwan

A new AI forecast model, GraphCast, was recently used to track Hurricane Beryl on July 1, a week before the landfall struck Houstonians in Houston, Texas. What once took hours to do relying on a supercomputer may now be completed within minutes with any laptop.

At the beginning of July, planes and spacecraft detected the birth of Hurricane Beryl around the Caribbean Sea. Weather experts immediately began experimentally predicting when the hurricane’s landfall would be in Texas with Artificial Intelligence, which concluded the same forecast predictions as a current model would, but faster and easier.

“This is a really exciting step,” Matthew Chantry, AI specialist at the European Center for Medium-Range Weather Forecasts, said, implying further development for AI weather technology.

Areas that experience common hurricanes, droughts, or other unusual weather patterns may find this new machinery beneficial. GraphCast also provides decision-making services to industrial workers, for example, when a farmer should irrigate his crops.

GraphCast can update weather data within a certain region in mere minutes, as opposed to conventional weather machines that need hours. The AI produces forecasts at a resolution of .25 degrees longitude/latitude, using over a million grids to cover the entirety of Earth’s surface. Within each grid, the model contains predicted data of temperature, wind speed and direction, humidity, and more. Rémi Lam, GraphCast’s lead scientist, said, “[GraphCast] learns from historical data.” The model can construct a ten-day forecast in seconds, labeling itself the most precise weather forecast in the world.

GraphCast and other Artificial Intelligence weather models will not completely take over meteorology because mistakes are very possible as unpredictable weather changes often occur and the AI might miscalculate.

AI researchers predict that robots will outperform humans within the following decade in activities such as driving and surgery. Until then, the direction of the continuously advancing GraphCast is vast.