Around this work lie studies of solar cycles, paleoclimate records, and El Niño forecasts up to a year and a half ahead. Neighboring studies build static networks or track a single measure, whereas here, for the first time, the evolution of several global network characteristics is traced simultaneously. This approach places the paper in the stream of searching for precursors of climate shifts, but with a richer toolkit than its predecessors.
The input data are daily surface air temperature anomalies from the NCEP/NCAR reanalysis over 62 years, regridded onto a nearly isotropic icosahedral grid of 10,242 nodes to avoid distortions from point clustering near the poles. For each sliding window of one year with a 30-day shift, pairwise Pearson correlations at zero lag were computed. Only the top 0.5% strongest links were included in the network, that is, about 262,000 edges out of more than 52 million possible. Then, for each window, four global characteristics were calculated: mean path length, transitivity, global clustering coefficient, and the Hamming distance between adjacent windows.
Transitivity and mean path length of the network rise sharply during strong eastern-type El Niño events, such as 1982/83 and 1997/98, while the global clustering coefficient drops. These surges coincide with the appearance of "localized structures," which are densely connected regions in the tropical Pacific. Central-type El Niños, such as 1990/91 and 2004/05, leave almost no trace in these characteristics. Volcanic eruptions, especially Pinatubo in 1991, produce a similar but delayed signal by several months. The Hamming distance between adjacent windows shows a double peak around eastern El Niños, reflecting the rewiring of links as structures emerge and vanish.
Unlike neighboring studies that use a single measure or a fixed threshold, here the simultaneous analysis of several characteristics allows functional distinction between El Niño types. This is the first demonstration that a climate network can separate eastern and central types without resorting to ocean surface temperature indices. The work also shows, for the first time, a noticeable network response to volcanic eruptions, extending the method's applicability beyond ENSO.
The approach can be extended to coupled climate networks that combine several observed variables, such as ocean temperature and atmospheric pressure. Further study of local structures and link anisotropy may reveal the physical mechanisms behind the difference between El Niño types. Possibly, network evolution could provide early signs of an upcoming event several months before its peak.
The method is useful for climatology, where it complements traditional principal component analysis, and for Earth sciences in general, where networks are already applied to hydrology and seismology.
Test whether surges in transitivity and Hamming distance can predict the type and strength of an approaching El Niño. Compare the network behavior with indices based on ocean surface temperature for independent classification of events.
The work addresses the open question of physical differences between eastern and central El Niño types. Climatology still lacks a unified classification of these events, and the network approach offers an independent view of their spatiotemporal organization.
The edge density in the network is only 0.5%, meaning that out of more than 52 million possible node pairs, only about 262,000 are connected. Yet the network still captures global climate events.