El Niño remains the main source of interannual climate variability, and its impacts, such as droughts and floods, are felt on every continent. Most current models lose predictive skill beyond the so-called "spring barrier," a period when the ocean and atmosphere are particularly unstable. Surrounding research either improves short-term forecasts or analyzes historical weather data, but none provides a systematic real-time test of long-term prediction. This paper fills that gap: it evaluates a method that was fixed in 2013 and has not changed since, ruling out tuning to known events.
The authors use a climate network built on 14 nodes inside the El Niño basin and 193 nodes outside it. For each pair of nodes, they compute the temporal cross-correlation of surface air temperature anomalies with lags up to 200 days. Link strength is defined as the ratio of the difference between the maximum and the mean correlation to its standard deviation, a signal-to-noise measure. The average strength of all links, S(t), is compared with a threshold Θ=2.82, which was chosen using data from 1950–1980. If S(t) crosses the threshold from below and the latest ONI index is below 0.5°C, an alert is issued for El Niño onset in the following calendar year.
From 2011 to 2022, the algorithm issued four alerts: in 2013, 2017, 2019, and 2022. The 2013 and 2017 alerts correctly predicted the events of 2014–2016 and 2018–2019, and the 2022 alert predicted the event that began in May 2023. The only false alarm was in 2019. In the other eight years, the algorithm correctly predicted no new event. Statistical significance was assessed via the p-value: the probability of obtaining the same or better forecast configuration by random guessing was about 0.005 for the real-time period and 3×10⁻⁵ for the entire period since 1981. This rejects the null hypothesis at a high significance level.
This work differs from neighboring studies in that it tests the method not on historical data but in real time over 12 years, when the future was unknown. This eliminates the "publication bias" inherent in retrospective forecasts. The approach has no free parameters after fixing the threshold, making it reproducible and transparent. The method's success indicates that the cooperativity of atmospheric processes in the Pacific a year before the event is indeed a physical precursor to El Niño.
The algorithm can be improved by adding a second step: if after a preliminary alert the ONI index does not remain below 0.5°C until the end of the year, the alert is canceled. This eliminates all false alarms in the considered period. Combining the network approach with other statistical methods to predict the magnitude and type of event is also promising. The question of forecasting La Niña, especially its duration, remains open.
The method could be used for early warning of El Niño in agriculture, water resource management, and disaster risk reduction planning.
Real-time monitoring should continue, and the improved algorithm should be tested on future events. The physical mechanism linking increased cooperativity to subsequent El Niño also warrants investigation.
For climatology, the "spring barrier" problem remains unsolved: a sharp drop in ENSO predictability during spring months. This method bypasses the barrier by using the slowly varying cooperativity of the atmosphere, which may point the way to a solution.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The study evaluates a climate network method for real-time El Niño forecasting over 2011–2022. It uses a network of 14 nodes within the El Niño basin and 193 nodes outside it, with link strengths computed from cross-correlations of temperature anomalies at lags up to 200 days. A threshold of S(t)=2.82, fixed on data from 1950–1980, yields four alarms over the 12 years, three of which are correct, and a p-value of about 0.005 rejects chance.
Where its strength is
Real-time validation over a 12-year horizon with a fixed threshold is a rare example of long-term testing without tuning to known events. The method's success points to physical predictability of El Niño through cooperative atmospheric processes up to a year ahead.
Where it could go next
Worth trying
Why this may matter
Long-lead El Niño forecasts a year in advance with high significance could improve preparedness for climate risks, especially in regions dependent on rainfall and ocean temperature. If the method proves robust, it could become a cheap and transparent early-warning tool.
Nearby works in our archive
The closest work in the archive sits at 0.70 by meaning — the usual level for our corpus.
The threshold Θ=2.82 was chosen from the range 2.815–2.834, which gave the best results on the training period 1950–1980. Interestingly, all false alarms (1994, 2004, 2019) were accompanied by at least one ONI value ≥0.5 in the same year, which inspired the idea for improving the algorithm.