Read our COVID-19 research and news.

The aftermath of an El Niño event in Peru in 2017


Artificial intelligence could predict El Niño up to 18 months in advance

The dreaded El Niño strikes the globe every 2 to 7 years. As warm waters in the tropical Pacific Ocean shift eastward and trade winds weaken, the weather pattern ripples through the atmosphere, causing drought in southern Africa, wildfires in South America, and flooding on North America’s Pacific coast. Climate scientists have struggled to predict El Niño events more than 1 year in advance, but artificial intelligence (AI) can now extend forecasts to 18 months, according to a new study.

The work could help people in threatened regions better prepare for droughts and floods, for example by choosing which crops to plant, says William Hsieh, a retired climate scientist in Victoria, Canada, who worked on early El Niño forecasts but who was not involved in the current study. Longer forecasts could have “large economic benefits,” he says.

Part of the problem with some El Niño forecasts is that they rely on a relatively small set of historical statistics for factors such as ocean temperature. Other forecasts use climate models but struggle to create the detailed pictures of the ocean needed for long-range forecasts.

The new research uses a type of AI called a convolutional neural network, which is adept at recognizing images. For example, the neural network can be trained to recognize cats in photos by identifying characteristics shared by all cats, such as whiskers and four legs. In this case, researchers trained the neural network on global images of historic sea surface temperatures and deep ocean temperatures to learn how they corresponded to the future emergence of El Niño events.

Such neural networks need a large number of training images before they can identify underlying patterns. To get around the shortage of historic El Niño data, the scientists fed the program re-creations of historic ocean conditions produced by a set of reputable climate models, ones frequently used for study climate change, says the study’s lead author, Yoo-Geun Ham, a climate scientist at Chonnam National University in Gwangju, South Korea. As a result, the scientists could show the computer system not just one set of actual historic data, spanning 1871 to 1973, but several thousand simulations of that same data by the climate models.

When tested against real data from 1984 to 2017, the program was able to predict El Niño states as far out as 18 months, the team reports today in Nature. The program was far from perfect: It was only about 74% accurate at predicting El Niño events 1.5 years into the future. But that’s still better than best current model, which is only 56% accurate for that time frame, Ham says.

The AI also proved more adept at pinpointing which part of the Pacific would heat up the most. That has real-world implications, because El Niños centered in the eastern Pacific, closer to South America, translate into hotter water temperatures in the northern Pacific and more flood-inducing rain in the Americas, compared with El Niños that are centered farther to the west.

The use of the climate models to create extra training data is a clever way around the shortcomings of other approaches, Hsieh says. It appears enough of an advance that it should be deployed for real forecasts, he says.

But it’s not clear how much real-world benefit will come from pushing forecasts beyond 1 year, cautions Stephen Zebiak, a climate scientist and El Niño modeling expert at Columbia University’s International Research Institute for Climate and Society in Palisades, New York. “The kind of lead time that is actionable is probably less than a year,” because decision-makers are unlikely to take action further in advance, he says.

The researchers have already begun to issue forecasts extending into 2021, and predict a likely La Niña event—El Niño’s cooler opposite—which can bring heavier monsoons and droughts. But major government forecasting agencies are not yet considering the group’s predictions. Ham says he and his colleagues are tweaking the model to extend the forecast even further. Meanwhile, he says his team is now working to improve forecasts for another ocean pattern, the Indian Ocean Dipole. That fluctuation in ocean temperatures can influence rain and tropical cyclones in Asia and Australia.