This year, the weather has messed up, and severe weather events have become the subject of disasters across the planet.
Unprecedented heat has gripped all the countries of the world, and the summer season that just ended was considered the hottest in history.
From floods in California and ice storms in Texas, to devastating wildfires in Maui, Greece or Canada, weather-related events have deeply affected communities and claimed lives, while also wiping out much of the flora and fauna.
See also: Artificial intelligence giants, more “secretive” than is good for mankind. What was found out
The weather, predicted by artificial intelligence from Google
Recently, Google DeepMind released an AI that provides 10-day weather forecasts with unprecedented accuracy and speed.
Called GraphCast, the model can analyze hundreds of weather data for a given location and generate predictions in less than a minute.
Although it wasn’t necessarily trained to detect extreme weather patterns, the AI identified several atmospheric events.
GraphCast also predicted the onset of extreme temperatures well ahead of traditional methods.
Unfortunately, 2024 is predicted to be even warmer and with increasing extreme weather events. Under these conditions, AI predictions could give communities precious time to prepare for impact.
See also: Romanians’ opinion about artificial intelligence. Why some consider it harmful
How GraphCast actually works
GraphCast is based on a type of neural network that has previously been used to predict other physics-based systems, such as fluid dynamics, and consists of a few essential parts.
First, the encoder maps the relevant information. You can think of this as an abstract infographic that machines can easily understand.
The second part is the processor, which learns to analyze and transmit information to the final part, the decoder. In turn, the decoder translates the results into a real-world weather forecast map.
In total, GraphCast can predict weather patterns for the next six hours.
Six hours, of course, is not ten days, but the AI can learn from its own experiences.
GraphCast forecasts are fed back into the AI, allowing it to progressively predict the weather further out in time.
GraphCast was trained on nearly four decades of historical weather data. The team divided the planet into smaller portions, and this helped the artificial intelligence understand how to analyze the local weather.
See also: Artificial intelligence has already “beaten” us to creativity. What scientists say about AI’s artistic capabilities
