Pacific’s Slow Climate Switch Rewires How Tropics and Poles Talk to Each Other

Deep in the Pacific Ocean, a slow-moving pendulum swings back and forth over decades, and according to new research, its position quietly reorganizes the entire wiring of the global climate system. A team of Australian scientists has shown that the phase of the Interdecadal Pacific Oscillation, or IPO, a multidecadal see-saw of sea surface temperatures spanning the whole Pacific basin, fundamentally changes how the tropics and the extratropics influence one another. The finding, published in the journal Nonlinear Processes in Geophysics, suggests that long-term background states of the ocean-atmosphere system are not passive backdrops but active gatekeepers that open and close the channels through which climate signals travel around the planet.

Mark Collier of CSIRO Environment in Melbourne, together with Dylan Harries of the South Australian Health and Medical Research Institute and Terence O’Kane of CSIRO Environment in Hobart, applied a technique known as Bayesian structure learning to decades of climate data. Rather than simply measuring correlations between climate patterns, their method infers directed networks of Granger-causal relationships, in which one time series is said to cause another if knowledge of its past improves predictions of the other. The result is a dynamic Bayesian network: a graph in which the major modes of climate variability appear as nodes, and the most probable lagged influences between them appear as arrows pointing forward in time. Because the approach is Bayesian, it does not deliver a single definitive diagram but a probability distribution over possible networks, allowing the researchers to quantify exactly how confident they are in each inferred connection.

The team drew on the ERA5 reanalysis, the European Centre for Medium-Range Weather Forecasts’ state-of-the-art reconstruction of the historical atmosphere, which blends observations with numerical weather prediction to provide a physically consistent record back to 1940. From this dataset they computed a battery of standard climate indices: the multivariate ENSO index capturing El Niño-Southern Oscillation variability in the tropical Pacific, the Indian Ocean Dipole measuring east-west sea surface temperature gradients in the Indian Ocean, the Madden-Julian Oscillation indices RMM1 and RMM2 tracking the eastward-propagating pulse of tropical convection, and a suite of extratropical atmospheric patterns including the North Atlantic Oscillation, the Pacific North American pattern, the Arctic Oscillation, the Southern Annular Mode, the Pacific South American patterns, and Scandinavian blocking. The IPO itself was represented by the Tripole Index, which contrasts sea surface temperature anomalies in the central tropical Pacific against those in the northern and southern subtropics.

The observational record poses a stubborn problem for anyone studying multidecadal variability: it contains only one clear positive IPO phase, from roughly 1977 to 1998, and one clear negative phase, from 1948 to 1976, long enough for reliable statistical fitting. To overcome this limitation, the researchers turned to an ensemble of historical simulations from the ACCESS-CM2 coupled climate model, run under CMIP6 protocols from 1850 to 2014. Because each simulation evolves its own internal variability, the ensemble provides multiple independent realizations of IPO phase transitions, giving the team a much richer sample of background states than the single observed transition. The trade-off was that daily model output needed to compute the MJO indices was unavailable for the model runs, so the researchers ran parallel analyses of the reanalysis data with and without the MJO to make fair comparisons.

The headline result is that the causal architecture of the climate system looks strikingly different depending on which way the IPO is leaning. During the negative IPO phase, which is characterized by cooler tropical Pacific waters, the IPO index shows strong autocorrelation persisting out to six-month lags, and the learned networks reveal an enhanced role for extratropical teleconnections acting on the tropics. Long-lagged influences from the Pacific North American pattern, the Southern Annular Mode, and the IPO itself become prominent. During the positive IPO phase, by contrast, the picture flips: IPO autocorrelation shrinks to about two months, while the Indian Ocean Dipole becomes far more persistent, and the Madden-Julian Oscillation emerges as a powerful conduit linking tropical convection to both the IOD and ENSO. This is consistent with the enhanced Walker circulation that prevails when tropical Pacific waters are warm, and with the observation that El Niño events are more frequent and longer-lived during positive IPO periods.

Perhaps the most striking discovery concerns the humble Madden-Julian Oscillation, an eastward-moving envelope of clouds and rain that circles the tropics every one to two months. When the researchers removed the MJO indices from their analysis, the learned networks underwent a dramatic collapse. Crucial pathways connecting the Northern Hemisphere’s tropospheric modes, including the North Atlantic Oscillation and the Pacific North American pattern, to equatorial sea surface temperatures vanished, while posterior edge weights between ENSO, the IPO, and the IOD swelled to compensate. In other words, the MJO is not merely an intraseasonal curiosity confined to the Maritime Continent; it is the key intermediary that transmits extratropical variability down into the tropical ocean-atmosphere system and helps sustain the very persistence of ENSO and the IPO themselves. Without it, the tropics and extratropics become largely strangers to one another in the inferred graphs.

The directionality of the learned edges also carried physical meaning that aligns with independent lines of evidence. The networks recovered a robust one-month-lagged influence of the negative phase of the North Atlantic Oscillation on the second MJO index, meaning that a strong negative NAO tends to precede and shape subsequent MJO behavior. This dovetails with earlier studies showing that seasonal prediction models initialize MJO forecasts more skillfully when the NAO state is strong and negative, and with proposed mechanisms involving extratropical Rossby wave propagation into the tropical upper atmosphere. Similarly, the analysis found that when the IPO is included as a node, the Southern Annular Mode exerts a statistically significant influence on ENSO at six-month lags, a relationship that paleoclimate reconstructions had hinted at but that the instrumental record had been too short to pin down.

Comparisons with the ACCESS-CM2 model ensemble revealed both the promise and the limits of current climate models. The simulated networks showed considerable diversity across the three IPO transition cases examined, including one in which a nineteen-year neutral period separated the negative and positive phases, and the extratropical-to-tropical connections were systematically stronger than in the reanalysis. The model also exhibited known biases, with an ENSO-like sea surface temperature pattern extending too far west and weakened anomalies over the tropical Indian Ocean. Notably, the observed IOD-to-ENSO connection present during the negative IPO phase did not appear in any of the modeled networks. These discrepancies matter, the authors argue, because the current generation of models often shows large systematic errors in autocorrelations and dependencies, and causal network diagnostics offer a sharp new tool for identifying exactly where those errors originate.

The implications reach well beyond academic diagram-making. The IPO has been linked to periods of accelerated global warming and to the early-century warming hiatus, to Australian rainfall and flood risk, and to the modulation of the Walker and Hadley circulations that shape weather across the Indo-Pacific region. If the strength and even the direction of teleconnections depend on the IPO phase, then seasonal-to-decadal prediction systems may need to condition their forecasts on the background state, and the intrinsic predictability of modes like ENSO may itself be regime-dependent. The authors caution that their directed graphs indicate the probability that causal relationships exist, and that verifying the underlying mechanisms requires detailed follow-up investigation. But the message is clear: the climate system’s wiring diagram is not fixed. It is redrawn, decade by decade, by the slow breathing of the Pacific.

Subject of Research: The influence of Interdecadal Pacific Oscillation phase on tropical-extratropical climate teleconnection dependencies

Article Title: Inferring the role of Interdecadal Pacific Oscillation phase on tropical-extratropical teleconnection dependencies

Article References: Collier, M. A., Harries, D., & O'Kane, T. J. (2026). Inferring the role of Interdecadal Pacific Oscillation phase on tropical-extratropical teleconnection dependencies. Nonlinear Processes in Geophysics, 33(1), 103-122. https://doi.org/10.5194/npg-33-103-2026

Image Credits: AI Generated

DOI: 10.5194/npg-33-103-2026

Keywords: Interdecadal Pacific Oscillation, ENSO, Madden-Julian Oscillation, Indian Ocean Dipole, teleconnections, Bayesian structure learning, causal inference, ERA5 reanalysis, climate variability, Walker circulation, North Atlantic Oscillation, CMIP6