El Niño–Southern Oscillation (ENSO) remains the main source of interannual climate variability. It swings precipitation and temperature across the globe, and strong El Niños against a warming background regularly push global temperatures to new highs. Operational forecasts based on dynamical models hit the 'spring predictability barrier': reliable warnings shrink to six months. This work joins attempts to overcome that barrier with statistical methods that look for precursors in data rather than in hydrodynamic equations. Nearby in the archive are papers on forecasting El Niño a year and a half ahead, on reviving old weather records, and on distinguishing noise from signal in climate series.
The first method builds a climate network from 14 nodes in the central and eastern equatorial Pacific and 193 nodes outside that region. Nodes are linked if their surface air temperature anomalies are strongly correlated. The mean link strength S(t) rises in the year before an El Niño onset and falls as it begins. The threshold Θ is fixed on the training period 1950–1980. The second method computes system sample entropy (SysSampEn) over 22 nodes in the Niño3.4 region at 5°×5° resolution. SysSampEn measures the complexity of temperature anomalies: low values indicate order, high values disorder. One year before El Niño, SysSampEn is typically high, while before a neutral year it is low. To discriminate La Niña from neutral, logistic regression on the ONI index from October–December of the previous year is used.
Both methods gave a consistent forecast: no El Niño in 2025, with 91.2% probability from the climate network and 91.7% from SysSampEn. The mean link strength in 2024 remained well below the threshold, and SysSampEn was 0.79 against a threshold of 1.31. On historical data since 1981, the strict version of the network method gave 9 correct alarms out of 9, and forecasts of no El Niño were correct in 31 out of 34 cases. Combining with logistic regression on ONI yields a 69.6% probability of a neutral phase and a 21.8% probability of La Niña for the November–January 2025/26 season. This is notably higher than the neutral probability given by the official NOAA CPC forecast (44% for August–October 2025).
The work shows that statistical precursors can look past the spring barrier up to a year ahead. Unlike neighboring studies that either predict only El Niño onset or only its magnitude, here two independent approaches are combined for the first time, with a probabilistic layer added to distinguish La Niña from neutral. This is not a revolution but a careful gain in accuracy: the forecasts align with historical patterns yet give higher confidence in a neutral outcome than dynamical models.
The methods could be extended to forecast the magnitude and type of El Niño (eastern Pacific or central Pacific), as has been done in previous work. Additional predictors might be included, such as the state of the Indian Ocean or stratospheric processes. The long-term goal is to turn these statistical approaches into an operational tool that complements dynamical models and helps smooth out the spring barrier.
A forecast of neutral ENSO in 2025 matters for agriculture, water management, and energy in regions sensitive to El Niño. It also suggests a slight drop in global temperature after the record year of 2024.
Verify the forecast against actual ONI data at the end of 2025. If successful, integrate the methods into routine monitoring and compare them with dynamical model ensembles over longer records.
An open question in climatology is the physical mechanism linking increased correlations in the tropical Pacific to a subsequent El Niño. Statistical precursors work, but their physical interpretation remains unclear. This is part of a broader problem: how to distinguish causal links from coincidental patterns in a complex climate system.
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 develops two statistical methods for ENSO forecasting: an analysis of a climate network of 14 nodes in the equatorial Pacific and the computation of sample entropy across 22 nodes in the Niño3.4 region. Both methods consistently predict no El Niño in 2025 with a probability of about 91%, while combining them with logistic regression on the ONI index yields a 69.6% probability of neutral conditions and a 21.8% probability of La Niña for the November–January 2025/26 season.
Where its strength is
The main strength is the combination of two independent statistical approaches, which boosts confidence in the forecast, and the addition of a probabilistic layer to distinguish between neutral conditions and La Niña, something missing from previous studies.
Where it could go next
Worth trying
Why this may matter
If statistical precursors can indeed look past the spring barrier a year ahead, this could improve operational ENSO forecasts and, in turn, seasonal climate predictions for many regions that depend on El Niño.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.52, against a usual 0.64. It looks like few have drilled nearby.
Since 1981, the strict version of the network algorithm has issued 9 El Niño warnings, and all 9 were correct. Not a single false alarm.