Forecasting the El Niño–Southern Oscillation (ENSO) remains one of the hardest problems in climatology. Traditional dynamical models hit the 'spring predictability barrier,' shrinking the reliable forecast horizon to about six months. In our archive, related studies also seek ways past this barrier: some use neural networks to predict droughts months ahead, others analyze solar cycles as a source of long-term climate memory, and still others separate weather noise from climate signal. This work stands somewhat apart because it relies not on physical modeling but on statistical patterns in the dynamics of temperature anomalies themselves.
The authors use two independent statistical approaches. The first builds a climate network from 14 nodes in the central and eastern equatorial Pacific and 193 nodes outside that region. Link strengths between nodes are set by cross-correlation of surface air temperature anomalies. When the average link strength S(t) crosses a threshold Θ from below, and the current ONI index stays below 0.5°C, a signal is issued for a possible El Niño in the following calendar year. The second method computes the systemic sample entropy (SysSampEn) of temperature time series in the Niño3.4 region. This measure of system complexity correlates with the magnitude of a future El Niño: the SysSampEn value from the previous year can estimate event strength via linear regression. The network uses NCEP/NCAR reanalysis data, while the entropy method uses ERA5.
The climate network based on NCEP data did not cross the threshold band during 2025, so it forecasts no El Niño in 2026 with 91.4% probability. The entropy method on ERA5 data gives SysSampEn = 1.60 against a threshold of 1.50, meaning a forecast of weak El Niño with magnitude 0.84 ± 0.36°C and 71.4% probability. The discrepancy partly stems from different data sources: applying the network algorithm to ERA5 shows a partial crossing of the lower threshold, which can be read as a weak signal. Over historical data from 1984 to 2025, there were six cases where the entropy method predicted El Niño but the network did not; in two of those (2004, 2006) the event did occur, and in four it did not. By Laplace's rule, this gives a probability of El Niño of (2+1)/(6+2) = 37.5%. If the event does occur, the index type points to an eastern Pacific type with 76.9% probability. If no El Niño occurs, logistic regression on the current ONI = −0.5°C gives 29.1% for La Niña and 70.9% for neutral conditions, which after multiplying by the probability of no El Niño yields unconditional probabilities of 18.2% and 44.3%, respectively.
The work shows that statistical methods based on network cooperativity and entropy complexity can provide forecasts with a horizon of about a year, overcoming the spring barrier. Unlike neighboring studies that either build neural network models on big data or analyze individual climate indices, here two physically distinct indicators are combined: large-scale atmospheric connectivity and local complexity of the temperature series. Their divergence is not dismissed as noise but used to quantify uncertainty via Laplace's rule. This is an honest way to show that a model does not always give a clear-cut answer, and that different data sources can lead to different conclusions.
Further development will likely formalize partial threshold crossings, as seen with the ERA5 version of the network. The authors explicitly note that a detailed probabilistic interpretation of such partial signals will be the subject of future work. One can also expect attempts to merge the network and entropy approaches into a unified Bayesian scheme that accounts for correlations between their errors, rather than simply applying Laplace's rule to historical discrepancies.
The results matter for seasonal climate forecasting, especially in ENSO-sensitive regions: agriculture, water resource management, and energy. Earlier and more honest probabilistic estimates allow better preparation for droughts or floods.
The immediate next step is to wait for actual developments in 2026 and compare them with the forecast to update the methods' success statistics. Then, researchers should investigate why the network method behaves differently on ERA5 than on NCEP, and whether a unified dataset can eliminate this discrepancy.
In statistical climate forecasting, the question of how to combine conflicting signals from different models remains unresolved. This work shows one possible path: using the historical frequency of discrepancies to calibrate probability. However, it is unclear how stable this frequency is over time and whether it changes with climate change.
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 authors apply two statistical approaches to forecast El Niño for 2026: a climate network based on cross-correlations of temperature anomalies and the systemic sample entropy of time series in the Niño3.4 region. The network indicates the absence of an event with a probability of 91.4%, while the entropy method predicts a weak El Niño with a magnitude of 0.84±0.36°C and a probability of 71.4%. Combining the discrepancies using Laplace's rule, the authors estimate the probability of El Niño at 37.5%, which is lower than the official NOAA forecast of 61%.
Where its strength is
The strength of the work lies in demonstrating how two independent statistical indicators, based on different properties of the climate system (network cooperativity and entropy complexity), can complement each other and reveal uncertainty in long-term forecasts, overcoming the spring predictability barrier.
Where it could go next
Worth trying
Why this may matter
If statistical methods relying on independent indicators can consistently provide forecasts with a horizon of about a year, this could substantially improve early warning for El Niño and associated extreme weather events, which is important for agriculture, water resource management, and disaster risk reduction.
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.
Laplace's rule, used to combine the forecasts, was formulated in the 18th century to estimate the probability of the sun rising. Here it is applied to climate events, highlighting the link between classical probability theory and modern climatology.