Imagine you have 55 old photo albums, one for each year. Each album holds 365 snapshots of rain, but in the past you only looked at the single wettest shot from every album. The other 364 snapshots gathered dust in a box. The new method pulls out every snapshot and lays them all out in order.

First, each weather station is tuned individually, like a piano tuner before a concert. You take all the days when rain crossed a certain threshold and build a curve from those days. The curve shows how likely an extreme downpour is. Before, that curve had only 55 points; now it has hundreds, so it comes out smoother.

Then the stations learn to listen to each other. If rain in two neighboring towns almost always starts at the same time, they likely share a common cause, such as a large atmospheric front. This connection is measured using the strongest rains of each year. Once the connections are tuned, the risk map comes into focus. For California, the new method detected El Niño's influence at 22 stations, while the old one found it at only 14. And the corridor of possible error narrowed by about a third.