Climate scientists have long seen that the temperature at one point on Earth is linked to the temperature at another, sometimes thousands of kilometers away. These links are called teleconnections, and they are especially strong during El Niño. Alongside this work in the archive are studies that predict El Niño a year and a half in advance, investigate solar cycles, and examine the ocean conveyor belt. All of them seek patterns in global climate data, but this article stands out because it looks not at individual indices but at the entire network of connections at once, like a living organism that reorganizes under stress.
The researchers took daily surface air temperatures from 1948 to 2015 on a grid of 10,242 points covering the Earth almost uniformly. For each sliding window of one year, they computed correlations between all pairs of points and kept only the 0.5% strongest links, turning them into network edges. They then measured three network properties: transitivity (how much links form dense clusters), modularity (how much the network splits into isolated communities), and average link length (how far edges typically reach). For volcanoes, they additionally excluded the broad El Niño region from the analysis to separate the effects of the two phenomena.
During strong El Niño and La Niña events, network modularity drops sharply while transitivity rises. This means the network loses its usual division into regional clusters and instead forms long-range connections spanning the entire globe. This is especially clear for the eastern Pacific type of El Niño: it creates teleconnections with Indonesia and West Africa that the central Pacific type does not. After the 1991 Pinatubo eruption, transitivity also jumped, but average link length fell, pointing to a local rather than global impact. Similar but weaker traces were found after the 1963 Agung eruption, while El Chichón in 1982 left no noticeable imprint on the network.
This work shows that the global climate system can be viewed as a network that responds differently to internal and external shocks. El Niño acts as a switch that temporarily erases boundaries between climate regions, while volcanoes, on the contrary, strengthen local connections. Unlike neighboring studies that focus on predicting El Niño or on individual indices, this one for the first time combines topological and spatial network measures, allowing a distinction between global reorganization and regional response. This is a new perspective on how climate transmits signals over distances.
The next step could be a quantitative check of how strongly modularity and average link length correlate with traditional indices such as ONI. It might be possible to create a new index that better distinguishes El Niño types or predicts their onset. It would also be interesting to test how eruptions of extratropical volcanoes, such as Mount St. Helens in 1980, which were not analyzed in this work, affect the network.
The method could be useful for real-time monitoring of climate anomalies and for validating climate models: if a model reproduces the observed network reorganization, it can be trusted more. It is also applicable to any spatially distributed data, from ocean currents to precipitation.
The authors plan to improve statistical significance by comparing network measures with existing indices and to study how volcanic aerosols affect monsoons through teleconnections. They also leave open the question of why, after 2000, regional network characteristics stopped returning to a low background level.
An unresolved question in climatology is exactly how El Niño transmits its signal to remote regions, especially Africa and Indonesia. This work shows that these teleconnections appear in the network structure but does not reveal the physical mechanism, such as the role of the Walker circulation. It is also unclear why different El Niño types produce different network responses and whether this can be used for forecasting.
The network was built from only the 0.5% strongest correlations, meaning that out of more than 52 million possible point pairs in each window, only about 260,000 links remained.