The Pacific Ocean spans half the planet, and its mood decides whether India gets rain or Australia gets drought. In the center of this giant, in a region that climatologists call NINO3.4, water temperatures rise and fall, affecting millions of lives. For a long time, the prevailing view was that the main conductor of this orchestra was the eastern part of the ocean, where the warm El Niño current is born off the coast of South America. But a new study using a neural network flips that picture: the western Pacific sets the rhythm, while the east merely plays along.
To understand why, imagine the ocean as a giant bathtub sloshing with waves. There are slow Rossby waves that crawl from east to west, and fast Kelvin waves that race along the equator. When wind pushes water off South America, these waves begin a journey across the entire ocean. Upon reaching the western boundary, Rossby waves reflect and return to the center as Kelvin waves. This return path, the authors argue, carries information about what will happen to temperatures in the central ocean months later.
In the experiment, the network learned from maps of surface temperatures over the previous 12 months and had to predict the NINO3.4 index for the next 12 months. The team trained three versions: one on the entire equatorial Pacific, another only on its eastern part, and a third only on the western part. The result was unexpected: the western part produced forecasts nearly as accurate as the full ocean, while the eastern part was noticeably worse. For lead times of two to five months, the west even outperformed the full model. This means that the temperature of the western part already encodes the fate of the central part.
The practical takeaway for forecasters is to pay closer attention to the western Pacific and perhaps include subsurface temperature data in models, especially the layer of sharp temperature change known as the thermocline. This work not only improves forecasting; it points to a physical mechanism: an asymmetry in ocean dynamics where the west acts as a reservoir of information about the future. So the Pacific orchestra finds its true conductor, and it stands on the western side of the stage.
The authors created 5000 versions of the same neural network, changing only the initial random filter weights. The spread in forecast accuracy among these versions was comparable to the difference between different architectures, echoing the sensitivity of chaotic systems to initial conditions.