The El Niño–Southern Oscillation (ENSO) remains the main source of interannual variability in the global climate, capable of causing droughts, floods, and economic losses across different parts of the planet. Existing operational forecast models typically fail to overcome the "spring predictability barrier," limiting the horizon to about six months. The work by Lüderscher, Meng, and Fan continues a line of research begun earlier at the Potsdam Institute for Climate Impact Research: the climate network method has already shown the ability to predict El Niño onset a year ahead, and the entropy method also predicts its strength. Unlike neighboring studies using machine learning or large-scale structure, this research relies on the physics of the climate system and requires no parameter tuning in the forecast phase.
The first method builds a climate network from 14 nodes in the equatorial Pacific (the "El Niño basin") and 193 nodes outside it. Nodes are characterized by surface air temperature, and the link strength between them is determined by cross-correlation. The key quantity is the average link strength S(t): its increase in the preceding year serves as a precursor to the event. A forecast is triggered when S(t) crosses the threshold Θ from below, and the current ONI index is below 0.5°C. The second method computes the system sample entropy (SysSampEn) from 22 time series of temperature anomalies in the Niño3.4 region at 5×5 degree resolution. The entropy of the previous year is linearly related to the amplitude of the future El Niño; a forecast is issued if the entropy exceeds the threshold of 1.35 and the December ONI is below 0.5°C. ERA5 reanalysis data from 1984 to 2021 are used, and the method parameters are fixed in advance and not tuned to the result.
In 2021, the network model did not record a threshold crossing, so it predicts no El Niño in 2022 with 90% probability. Over the period of real forecasts from 2012 to 2021, the method gave 9 correct predictions out of 10, which is statistically significantly better than random guessing (p≈0.022). The entropy method, in contrast, produced a high SysSampEn value of 1.95 for 2021, corresponding to a forecast amplitude of 1.90±0.23°C for 2022 with 69% probability. In historical data, there were 13 cases with high entropy and low December ONI, and in 9 of those cases El Niño indeed began. The authors explain the forecast discrepancy by noting that in 6 analogous situations (1984–2021), the network model was more often correct: in 4 of the 6 cases, El Niño did not occur.
This work demonstrates that two independent physical approaches can yield opposite forecasts and offers a way to resolve the conflict through historical analogs. Unlike neighboring studies that use machine learning or statistical models, here the forecast relies on explicit physical mechanisms: cooperative behavior of atmospheric links and the complexity of temperature series. This allows not only prediction but also interpretation of the result, which is important for trust in early warnings. The article also highlights the limitations of each method: the network approach does not provide amplitude, while the entropy method more often errs in cases where the network model remains silent.
Both methods could be improved by combining them: for example, if the network model gives no signal while the entropy method predicts a strong event, one could apply a correction based on the historical reliability of each approach in such situations. The climate network could be extended to other regions or include additional variables (pressure, wind). For the entropy method, testing on longer series and optimizing parameters for different El Niño types (eastern Pacific and central Pacific) is promising.
El Niño forecasts directly affect agriculture, water resource management, energy, and insurance in ENSO-sensitive regions: Southeast Asia, Australia, South America, and India. Early warning a year in advance allows preparation for droughts or floods.
Check the actual 2022 data and compare them with both forecasts. Conduct a retrospective analysis of all cases where the methods diverged to refine probabilistic estimates.
In climatology, the "spring predictability barrier" of ENSO remains an unresolved problem: most models lose skill when forecasting across spring. This work shows that methods based on network cooperativity and entropy can overcome this barrier, but their reliability is still below desired levels. The physical nature of the precursors remains an open question: why does the increase in correlations or complexity occur exactly one year before the event, and how is it related to ocean dynamics?
In 2012, when most forecasting centers expected El Niño with 75% probability, the climate network method had already predicted its absence in late 2011—and turned out to be right. This was one of the first public successes of the network approach.