For decades, scientists have worked to improve predictions of El Niño-Southern Oscillation (ENSO), a climate powerhouse that can cause droughts, flooding, marine heatwaves, and more around the world. Researchers from the University of Hawai'i at Mānoa recently published a study showing that they can skillfully predict El Niño and La Niña 15 months ahead of time using only observations of the ocean surface temperature and height–no complex climate model needed.
"Many of today's leading forecast systems are either computationally expensive dynamical climate models, statistical models that rely on ENSO knowledge built over decades of research, or AI approaches that require large amounts of training data and are often harder to interpret physically," said Yuxin Wang, lead author of the study and postdoctoral researcher with the UH Sea Level Center in the UH Mānoa School of Ocean and Earth Science and Technology (SOEST). "Our simpler, data-driven empirical climate model, built only from ocean observations related to two core climate memories known for over 50 years, achieves ENSO forecast skill comparable to, and in some cases better than, many of today's more complex climate models and leading AI-based approaches."
Building on past discoveries
Klaus Wyrtki, a pioneering oceanographer at SOEST in the 1960s through 1990s, was the first to show that sea level changes can reveal heat build-up in the tropical Pacific, which led him to propose using tide gauge observations to predict El Niño. Klaus Hasselmann, a German oceanographer and Nobel laureate, showed that the ocean can retain a memory of past climate conditions through large-scale temperature patterns, including sea surface temperature patterns outside the tropical Pacific that can still influence ENSO.
Building on these two principles, the SOEST team developed the "Wyrtki-CSLIM" computer model to predict ENSO, which they named in honor of Wyrtki's insights that eventually motivated creating the UH Sea Level Center to operate tide gauges around the world. Today, these same tide gauges are used to calibrate satellite measurements of the sea surface height, which made this study possible.
The team trained their model using ocean observations that correspond directly to the two forms of climate memory originally suggested by Wyrtki and Hasselmann. The first is sea level in the equatorial Pacific, which reflects heat stored in the upper-ocean and represents "Wyrtki memory". The second is global sea surface temperature, which captures "Hasselmann memory", or the lingering influence of faraway temperature anomalies that can contribute to El Niño or La Niña developing months later.
The researchers tested how well this model could predict the Niño3.4 index, a standard measure used to track El Niño and La Niña. To do that, they ran the model on six decades of past climate conditions and asked how well it would have predicted what happened next in the real world.
"We found that it can predict El Niño and La Niña surprisingly well, with useful skill up to about 15 months ahead," said Wang. "Accurately predicting ENSO more than a year in advance is important because it can provide early warning, allowing communities, governments, and resource managers to take actions and make adaptations to reduce the potential impacts from El Niño and La Niña."
Predicting future ENSO
The Wyrtki-CSLIM currently predicts the development of a strong El Niño, more than 2 °C warmer than normal over the equatorial eastern Pacific, toward the end of this year. This up-to-date ENSO forecast is available online at the UH Sea Level Center.
"Our Wyrtki model is predicting a stronger El Niño than most of the other statistical models, and it is in line with the much more sophisticated dynamical models," said Matthew Widlansky, study co-author and associate director of the UH Sea Level Center. "However, it is important to note that all models have uncertainties, and the climate impacts of each El Niño event are different."
This new research also offers a clear direction for other ENSO forecasting systems.
"Models aiming to predict ENSO well should represent these two types of climate memory accurately," Wang noted. "Importantly, this can be achieved with a model that is relatively simple, explainable, and low-cost. That suggests that capturing the key sources of ENSO predictability do not always require the computationally expensive models."
