El Niño–Southern Oscillation (ENSO) is the strongest natural driver of interannual global climate variability. It causes extreme weather conditions in various regions worldwide: droughts, floods, and anomalous heat. Existing operational forecasts have limited lead time, typically not overcoming the so-called 'spring barrier' and providing reliable warnings only 6 months ahead. In the archive, related works cover topics such as forecasting El Niño a year and a half in advance, seasonal space weather, historical weather records, precursor climate signals, neutrino oscillations, and solar cycles. This work stands apart because it focuses on two specific methods for forecasting El Niño with a lead time of about a year, which is notably longer than most neighboring studies.
The authors used two independent approaches. The first is based on climate network analysis: 14 nodes in the central and eastern equatorial Pacific are linked to 193 nodes outside this region. Link strengths are determined by cross-correlation of surface air temperature anomalies. The average link strength S(t) increases during the year before El Niño onset and decreases with its arrival. The threshold Θ was fixed during the training phase of 1950–1980. The second approach uses system sample entropy (SysSampEn), a measure of complexity (disorder) in time series of temperature anomalies in the Niño3.4 region. SysSampEn is computed from 22 grid points at 5°×5° resolution using daily surface temperature data (1000 hPa) from the ERA5 reanalysis. The SysSampEn value for the preceding year strongly correlates (r=0.90) with the magnitude of the upcoming El Niño, allowing a linear regression to forecast event strength.
Both methods predicted the onset of El Niño in 2023. In the past, there were 7 cases where both approaches gave a consistent onset prediction, and all 7 were correct. Applying a Bayesian estimate with Laplace's rule, the authors obtained a probability of El Niño onset of (7+1)/(7+2) ≈ 89%. The climate network method issued an alarm between May 24 and June 7, 2022, when the average link strength crossed the threshold from below, and the latest ONI index was below 0.5°C (-1.1°C for March-April-May). In the stricter version of the algorithm, requiring ONI to remain below 0.5°C until the end of the year, the condition was also met, as ONI was negative throughout 2022. The SysSampEn method gave a value of 1.78 for 2022, which exceeds the threshold of 1.235, and predicted an El Niño magnitude of 1.49±0.37°C. The correlation coefficient between SysSampEn of the previous year and the maximum El Niño magnitude is r=0.82, with a forecast RMSE of ±0.37°C. El Niño is expected to raise global temperature by about 0.2°C, which could make 2024 the hottest year on record. The probability of temporarily exceeding the +1.5°C threshold by the end of 2024 is estimated at 21.1%.
This work differs from neighboring archive entries by combining two independent El Niño forecasting methods with a lead time of about a year, whereas most neighboring studies either consider shorter lead times or use a single method. It is the first test of consistency between the two approaches on a single event, enhancing forecast reliability. Additionally, the study provides a quantitative estimate of the magnitude of a future El Niño, something many other statistical models do not do. The results highlight the possibility of overcoming the 'spring barrier' and providing early warnings of climate risks.
This line of research could advance toward refining forecasts of El Niño magnitude and type (central Pacific or eastern Pacific). It may also incorporate additional data, such as ocean surface temperature, salinity, and wind, to improve accuracy. Applying machine learning methods to automatically tune thresholds and parameters is also promising. In the future, an operational early warning system combining several independent indicators could be developed.
The results are important for climatology, meteorology, agriculture, water resource management, and planning adaptation measures for extreme weather events.
Next steps include validating the forecast on new data, refining model parameters, and developing an operational prototype of a warning system.
This work addresses the unresolved problem of ENSO predictability beyond the 'spring barrier' and the challenge of estimating event magnitude, not just its onset. It also touches on the relationship between the complexity of the climate system and the emergence of large-scale anomalies.
From 1981 to 2022, the climate network model issued 11 alarms, of which 8 were correct, and in the stricter version, all 8 alarms were correct.