El Niño affects temperature, precipitation, and atmospheric circulation over the Americas, Australia, Europe, India, and East Asia, and strong events trigger cascades of consequences for human livelihoods. Existing forecast models still do not provide reliable predictions of El Niño onset and especially strength with lead times beyond six months. The work aligns with studies that use climate networks to forecast El Niño, solar cycles, and other climate patterns, but differs in that it is the first to apply a disorder index within the Niño 3.4 region itself to simultaneously predict both the onset and the strength of the event.
The authors built a climate network of 105 nodes (grid points) in the Niño 3.4 region (5°S–5°N, 120°W–170°W) based on daily temperature anomalies. For each pair of nodes, cross-correlation was computed with time lags up to 200 days, and the link weight was determined by the maximum peak of this function. Then the forecast index FI was calculated as the difference of logarithms of the average network connectivity for the current and past months. An increase in FI means a decrease in coherence (increase in disorder) in the region. The forecast algorithm: if FI continuously rises for at least two months from a local minimum and crosses zero during a neutral period (ONI between -0.5 and 0.5), this signals a possible El Niño within the next 18 months. The event strength is estimated from the maximum peak of FI since the end of the previous El Niño. The method was tested on four datasets: ERA-Interim, NCEP Reanalysis I, PCMDI-AMIP 1.1.2, and ERSST.v5.
The FI index showed high accuracy in predicting El Niño onset: the hit rate ranged from 0.81 to 0.95 depending on the dataset, with false alarm rates from 0.09 to 0.24. The average lead time ranged from 6.9 to 12.2 months. For predicting event strength, the Pearson correlation between predicted and observed El Niño magnitude was 0.76 for ERA-Interim, 0.39 for NCEP Reanalysis I, 0.70 for PCMDI-AMIP, and 0.32 for ERSST.v5. The Kolmogorov-Smirnov test confirmed the statistical significance of the link between predicted and observed strength (D ≥ 0.37). Analysis of correlation distributions showed that the region becomes less coherent as it approaches the peak of disorder, then synchronizes with the onset of El Niño.
The work shows that a simple index based solely on temperature variability within the Niño 3.4 region can simultaneously predict the onset and strength of El Niño with a lead time of about a year. This differs from adjacent studies that either predict only onset or use more complex models. The method does not require data from other regions or ocean currents, making it computationally light and applicable to long observational records. The results point to an important characteristic of the ENSO cycle phase: a significant increase in disorder in the region well before the event begins.
Further improvement of the algorithm is possible by combining it with other forecast methods and models. The authors propose defining an entropy based on the coherence of temperature variations, which may oscillate periodically with the ENSO cycle. Applying the index to longer data records and other climate phenomena where spatial synchronization of processes is important also holds promise.
The method could improve early warning for El Niño for agriculture, water resource management, and disaster risk reduction in affected regions.
Test the index on independent data and in real time, and investigate the physical mechanisms linking the rise in disorder in the Niño 3.4 region to the subsequent El Niño.
The work addresses the fundamental problem of predictability of the climate system on interannual timescales. It shows that the internal spatial coherence of a key region can serve as a predictor of a large-scale phenomenon, opening the way to simpler and more reliable forecasting methods.
The FI index uses only temperature data from a small region of the Pacific Ocean, yet it predicts an event that affects weather worldwide. The strongest El Niño since 1948 (2014–2016) was predicted by this method a year before it began.