New AI techniques enable weather forecasts up to 33 days in advance, a significant improvement over the previous two-week limit, according to research from the University of Washington.

Until now, it was only possible to predict the weather up to two weeks in advance. However, thanks to new AI techniques, meteorologists will soon be able to forecast up to 33 days ahead. This is according to new research from the University of Washington.
The American scientists specifically studied GraphCast, an AI system from Google trained with 40 years of weather data and satellite images. The researchers improved the system's accuracy by refining the initial data it uses.
Previously, meteorologists believed that two weeks was the limit for accurate forecasts due to the butterfly effect: small events, like a gust of wind or a wildfire, can have significant long-term weather impacts. The further ahead you predict, the harder it becomes to account for these small changes.
The researchers addressed this issue by analyzing errors in the AI system's predictions and used this information to refine the initial data. This process was repeated over a thousand times. The result: 10-day forecasts became 86% more accurate on average. Even more impressive, the system could make predictions up to 33 days ahead.
While the researchers are excited, they caution that the system still needs to prove itself in real-world applications. Many more tests are required before this technology can be used for daily weather forecasts.