Climate networks have become a standard tool for analyzing complex climate phenomena such as El Niño, one of the most powerful climate phenomena on decadal timescales. El Niño warms the eastern tropical Pacific by several degrees, and the subsequent cooling is called La Niña. This phenomenon affects remote regions, including North America, Australia, Europe, the South China Sea, the Indian Ocean, and the tropical North Atlantic, leading to warming, increased rainfall in some regions, droughts in others, as well as fishery declines, famine, epidemics, political instability, and economic changes. The work sits at the fringe of the archive: nearby studies cover solar cycles, historical weather records, predicting El Niño a year and a half ahead, neural networks for Earth analysis, and ecosystems under nature's rhythm. These studies align with applying network and statistical methods to climate data, but this research stands out for its directed approach, focusing only on outgoing links from the El Niño basin.
The climate network is built from global daily surface air temperature fields (1000 hPa) from the NCEP/NCAR reanalysis dataset, with a spatial resolution of 2.5°×2.5° (10512 grid nodes in total), spanning January 1948 to April 2016. To remove seasonality, the mean seasonal cycle was subtracted from each time series and divided by the seasonal standard deviation. Nodes were split into two subsets: 57 nodes inside the El Niño basin (ENB, region 5°N–5°S, 120°–170°W) and 10455 nodes outside it. For each pair of nodes from different subsets, the cross-correlation of time series of length 365 days was computed with lags up to 200 days. The peak absolute cross-correlation and its corresponding time lag were found. The sign of the lag determines the direction of the link: if the lag is positive, the link originates from the ENB. Only outgoing links from the ENB with absolute lag no more than 150 days were considered. Link weight was defined as the normalized cross-correlation. For each node outside the ENB, the total 'incoming' weight from all ENB nodes was computed. El Niño and La Niña events were selected using the ONI index (3-month average sea surface temperature anomaly in the Niño 3.4 region) with a threshold of |ONI|>1°C. Eleven El Niño and nine La Niña events were considered. For each event, the time series from July 1 of the preceding year to June 30 of the following year was used. To compare events, heat maps of cross-correlations between 'incoming' weight maps were constructed, and significance was determined by shuffling the maps (dividing into 18 equal regions and permuting their order). Event communities were identified using a modularity optimization algorithm.
During El Niño and La Niña, regions with high 'incoming' weights coincide with known temperature anomaly areas: North America, South America, Europe, India, South Africa, and Australia. The cross-correlation between 'incoming' weight maps and mean winter temperature anomaly maps was 0.63 for El Niño and -0.59 for La Niña (for the measure based on C), indicating consistency. The key result: during El Niño/La Niña, the total area affected shrinks, but in these localized regions the impact intensifies. This is evident from comparing distributions of 'incoming' degrees: in El Niño years there are more black areas (no links) and more dark red areas (all 57 links from ENB incoming) than in normal years. Quantitatively, the number of nodes with nonzero 'incoming' degree drops sharply during events, while the average 'incoming' weight per node increases significantly. The reason is that during El Niño/La Niña, all 57 ENB nodes are synchronized (warm or cold simultaneously), so for each affected node outside the ENB, all 57 links have the same direction. In normal years, part of the ENB correlates and part does not, which reduces overall correlation and broadens correlation areas. Similarity analysis of events: the 11 El Niño events split into three communities based on significant cross-correlations of 'incoming' weight maps. The heat map for the tropics (20°S–20°N) is most similar to the global one, indicating the dominant contribution of tropical impacts to event classification.
This work is the first to use only 'incoming' directed links originating from the El Niño basin, allowing a map of global impacts for a specific event and the study of local effects. This differs from neighboring studies, which often use thresholds for significant links or analyze the entire network. The method reveals that El Niño's impact is not globally uniform but concentrates in limited regions, and these regions change from event to event. Event similarity, assessed through network community structure, offers a new way to classify El Niño types, complementing classification by the location of maximum sea surface temperature anomalies.
The approach can be applied to other climate variables, such as precipitation, pressure, or ocean surface temperature, to identify their teleconnections. Further development may include using longer data records, higher resolution, and other reanalyses to test the robustness of the results. It is also possible to include additional nodes in the El Niño basin or account for nonlinear interactions. Exploring the predictive power of 'incoming' weights for forecasting regional anomalies several months ahead is also promising.
The results could improve understanding and prediction of El Niño and La Niña events and their regional impacts. The method is applicable in climatology, meteorology, and related fields where teleconnections are important.
Verify reproducibility on other datasets (e.g., ERA-Interim) and for other climate variables. Investigate whether 'incoming' weights can be used for operational forecasting of anomalies in specific regions.
The work links climate network structure to the known problem of El Niño's heterogeneous impact: why some regions respond strongly while others do not. This helps refine teleconnection mechanisms and may contribute to solving the problem of predictability of regional climate anomalies.
During the strong El Niño of 2015–2016, one of the strongest since 1948, researchers were able to confirm most of the affected regions using data from previous events only, demonstrating the method's predictive potential.
In Kim Stanley Robinson's novel 'Forty Signs of Rain', climate networks and chaotic ocean dynamics play a key role in understanding abrupt climate change, echoing attempts to capture El Niño teleconnections.