AI Models Global Clouds, Saving Energy 50x ✛ ✦
STRATA works like an artist learning to paint the sky. Instead of an easel, a server; instead of studies, short video clips of clouds and storms from different ends of the Earth. The program soaks up these frames and then creates a holistic picture of the atmosphere over the entire planet for the next day.
Unlike old numerical models — cumbersome calculations where the machine, like a pedantic accountant, ticked off temperature, wind, and humidity for each point — STRATA sees the overall pattern. It captures regularities directly from the data, and this requires 50 times less energy.
The secret is that the program divides the sky into square “brushstrokes” and assembles them into a mosaic, considering only two key forces: solar heat and Earth's gravity. Even entropy — meteorological chaos — is not a hindrance, but a rhythm that can be learned. Amazingly, it took just 17 days of observations to master planetary clouds. As a result, a regular computer suffices instead of a power plant, and carbon dioxide emissions drop. Climatologists gain a fast tool for accurate predictions of downpours and hurricanes.
STRATA needed just 17 days of weather data — less than a typical vacation — to learn to predict clouds across the entire planet for the next day.
This model feels like a step toward the predictive machine from Isaac Asimov's science fiction. But now it's not fiction — it's a working tool for climate scientists.