An interpreter at negotiations hears only the words already spoken, yet still guesses the next phrase. This is how the new El Niño forecasting method works: it doesn't peek into the future but relies solely on the past, like an honest listener.
El Niño–Southern Oscillation (ENSO) is the dominant interannual climate fluctuation in the tropical Pacific Ocean. It causes droughts in Australia, floods in Peru, and alters weather worldwide. Forecasting it is difficult: in a single temperature series, fast noise, seasonal cycles, and slow decadal trends are mixed. Conventional smoothing methods that remove extraneous components secretly use future values of the series. For operational forecasting, this is unacceptable, because the future hasn't happened yet.
The filtered series was fed into a reservoir computer, a neural network with random internal connections where only the output layer is trained. The input vector consisted of nine values of the series with a delay step of 4 months. Network hyperparameters, such as reservoir size 244 and spectral radius 0.712, were tuned on the training period from 1986 to 1995. On an independent test from January 2001 to December 2015, the average correlation remains above 0.5 for 29 months. Accounting for the five-month shift, this means a real prediction horizon of 24 months, i.e., two years.
The model reproduces the major ENSO peaks in 2010–2014 well but underestimates the record El Niño of 2015–2016. Nevertheless, a simple reservoir computer with a properly chosen filter achieves a horizon comparable to more complex deep neural networks. The main achievement is that the entire pipeline from raw data to prediction operates strictly without future information.
An interpreter who doesn't peek at the speaker's text still guesses the next phrase from context. The new filter does the same for climate data: it hears only the past but speaks about the future two years ahead. Such an approach is applicable not only to El Niño but also to any slowly varying series, from financial to biomedical signals.
The standard Niño-3.4 index used to declare El Niño is itself smoothed with a five-month moving average, meaning the official status of the event at any given time already contains information about future months. The authors even abandoned that to make the forecast honestly operational.
In Kim Stanley Robinson's novel 'Forty Signs of Rain', scientists try to predict abrupt climate changes, grappling with chaos and incomplete data. The idea of a filter that works only with the past resonates with the dream of a forecast that does not depend on knowledge of the future.