Forecasts of the Pacific's great climate oscillation can be trusted up to a year in advance in some circumstances, according to research that examined more than 7,500 historical predictions from climate models built in nine countries.
The study, "Using initial states to identify successful El Nino-Southern Oscillation forecasts 1 year in advance", was published in Science Advances. It was reported in a University of Melbourne release carried by Phys.org.
Its lead author is Nandini Ramesh, a senior research scientist at CSIRO and an adjunct research fellow at UNSW Sydney. Mandy Freund of the University of Melbourne, a researcher with the Australian Research Council Centre of Excellence for 21st Century Weather, is a co-author.
El Nino and La Nina are the warm and cool phases of the El Nino-Southern Oscillation, which shifts rainfall and temperature across much of the planet. Events typically peak around December, and forecasters are usually confident about six months ahead.
The researchers found that the state of the Pacific at the moment a forecast begins - El Nino, La Nina or neutral - indicates how reliable that forecast will prove a year later.
Forecasts that began during La Nina performed well at one-year lead times. "The finding that one-year forecasts starting from La Nina perform well in all nine of the models tested suggests there is something about La Nina conditions that makes the tropical Pacific system more predictable over the following year," Freund said.
Forecasts that began from neutral conditions were far less reliable. Traditional physics-based models struggled with them, and the authors suggest machine learning and statistical approaches may do better.
"We've known for some time that the starting conditions affect how well model forecasts perform, but our study really shows how drastic those differences in performance can be," Ramesh said.
The approach is not ready for operational use. "We are still a few steps away from applying our findings to the models used in practice by forecasting agencies," Ramesh said.
The practical value of the result deserves to be stated plainly. A forecaster can now tell, at the moment a one-year forecast is issued, whether it is the reliable kind. The test is what the Pacific is already doing.
This gives the discipline something it has not previously had: a forecast about the forecast, available a full year before the thing being forecast.
The arrangement is internally sound. The Pacific is predictable when it is in La Nina, and whether it is in La Nina is a question about the present, which requires no prediction at all.
The Department of Unnecessary Research notes the single case the method cannot help with, which is the neutral Pacific: the ocean that has committed to nothing. Here the models struggle, and here the authors recommend other tools.
The finding can be stated as one rule, and this newspaper proposes it be adopted as one. The ocean can be read a year ahead whenever it has already made up its mind.