Did you know that an AI model outperformed a reference system in tests? In two thousand twenty-four, researchers present GenCast. The number counts, but you have to know what you compare: forecast performance, not certainty on each rainy day. The system produces several possible scenarios, to represent uncertainty. Our animated globe explains this idea: the colored areas remain an illustration. These are neither the real predictions of the day nor a map where each generated cloud corresponds to a phenomenon observed on Earth.
GenCast outperforms the European centre’s ENS system in 97.2% of the 1,320 comparisons. The evaluation uses 2019 data. This percentage is not a forecast accuracy rate and does not mean that weather forecasting is solved.
The interest is to compare methods with precise measurements and represent several possible futures. A forecast remains probabilistic. To organize an outing, it is necessary to consult a current weather service; this animation tells a dated result, without pretending to be a bulletin.
What we know
97.2% concerns1,320 comparisons with ENS, especially in 2019. This percentage is not a universal probability of correctly predicting the weather.
To go to the source
References consulted on October 12, 2026.
Signaler une erreur ou apporter une précision
