Every year, a different number of hurricanes is born in the Atlantic: sometimes five, sometimes fifteen. Scientists have long tried to figure out whether this number can be predicted in advance, using three main factors: warm ocean water, the El Niño phenomenon, and atmospheric pressure differences. These three things explain about half of why one year differs from another. But recently it turned out that further improvement of forecasts is no longer possible.

Scientists tested this on data spanning 140 years. They took a simple model with three factors and compared it with very complex methods that use maps of the entire ocean's temperature. It turned out that both the simple and the complex models make errors of about the same size: roughly 2.5 hurricanes. This means that extra complexity does not help. The reason is that hurricanes themselves arise randomly, like rare events, and this randomness cannot be circumvented by any computation.

Now researchers are thinking about where to go next. Perhaps there are hidden connections between neighboring years that have not yet been noticed. Or we need to abandon the assumption that hurricanes occur independently of each other. But so far, all checks show that randomness is indeed random, and the ceiling of accuracy has been reached. For people living on the coast, this is important knowledge: it is better to prepare for variability than to wait for a miracle from computers.