Imagine finding a trail of footprints on the beach. From them, you can tell whether the animal walked calmly or ran, and even predict where it will head next. Scientists did something similar with El Niño, the warm Pacific Ocean current that affects weather across the globe. Instead of building a massive ocean-atmosphere model, they took 135 years of observational records and tried to read the rule of its motion from them.
First, they filtered out rapid fluctuations, leaving only slow changes. Then they represented each moment as a point in a space where the coordinates are the record values on different days. It turned out that all points lie almost on a plane, not scattered randomly. This means there is a simple rule among the values, which can be written as a single formula. The formula is nonlinear: it accounts for the fact that warm and cold phases of El Niño behave differently.
Previously, El Niño forecasts deteriorated sharply every spring, a phenomenon known as the spring barrier. The new rule greatly weakens this barrier. This means we can now prepare in advance for droughts, floods, and other consequences of El Niño. And it all runs on an ordinary computer in seconds, not a supercomputer.
The model is trained on just 486 points (40.5 years of data) but forecasts 94.5 years ahead. Moreover, the computational cost of forecasting is practically zero, unlike the supercomputer calculations of physical models.