Introduction

Regime shifts are large, sudden, and persistent changes in the function and structure of natural systems1,2,3,4. Those occurring in the climate system are referred to as climate regime shifts (CRSs). CRSs manifest across various time scales, ranging from dramatic transitions between glacial and interglacial states to more abrupt multi-year to decadal shifts, which are hard to reverse and, in some cases, irreversible. Recent studies underscore the critical importance of abrupt climate system shifts—particularly those unfolding on multi-year to decadal timescales—which can profoundly reshape marine and terrestrial ecosystems, disrupt agricultural productivity, and threaten water security through large-scale alterations in weather patterns and storm tracks5,6,7,8,9,10,11,12,13,14. There is growing concern that global warming may increase the propensity of the climate system to undergo regime shifts15,16,17,18,19,20. Therefore, understanding and predicting the CRSs is essential for foreseeing their occurrences and implementing effective ecosystem management and climate adaptation strategies.

CRSs are typically triggered by strong disturbances, such as extreme events, or gradual changes in the underlying drivers and internal feedbacks3,4. Extreme events, including heatwaves, storms, droughts, and heavy rainfall, may easily push the climate systems beyond critical thresholds, leading to the occurrence of CRSs21. Meanwhile, gradual changes in the underlying drivers and internal feedbacks may progressively move the climate system closer to the critical threshold and trigger CRSs3,22. Additionally, stochastic noise, such as synoptic-scale atmospheric perturbations, can integrate into low-frequency variability that resembles a regime shift on multiyear to decadal scales23,24. Other mechanisms, including external forcings such as solar irradiance and volcanic activity, may also play a role in driving CRSs25,26.

Beyond these processes, large-scale climate modes represent a particularly important driver of CRSs, as they reorganize atmospheric teleconnections and amplify regional climate extremes. Among them, the El Niño-Southern Oscillation (ENSO), one of the most profound climate phenomena27,28,29, can serve as a key trigger of CRSs. Previous studies have demonstrated that ENSO events can induce regime-like transitions, including shifts in Pacific decadal variability and large-scale atmospheric circulations30,31,32,33. ENSO has also been linked to abrupt ecological transitions, ranging from coral bleaching and fisheries collapse in marine environments to vegetation dieback and hydrological regime shifts in terrestrial systems34,35,36,37. These impacts illustrate that ENSO can impose persistent climatic imprints extending well beyond its immediate phase, raising the question of whether its most extreme manifestations—super El Niño events—can trigger CRSs on a global scale.

Over the past few decades, several super El Niño events have occurred38,39,40, each exerting profound worldwide climate effects and substantially influencing ecological and socio-economic systems41,42. These intense air-sea coupling events disrupt atmospheric and oceanic circulations43,44, thereby altering global temperature and precipitation distributions. Although their immediate impacts typically persist for several seasons, it remains unclear whether these powerful anomalies can produce longer-term climatic legacies. They may potentially drive the system into a new state through positive feedback loops or the persistence of climatic memory.

While future frequency changes in super El Niño events remain uncertain45,46,47, their climate impacts are projected to intensify under continued global warming, with tropical and extratropical ENSO responses shifting eastward45,48,49. Within this context, two critical questions emerge: (1) Do super El Niño events increase the likelihood of CRSs? (2) How will super El Niño-induced CRSs respond to future global warming? Current understanding of these scientific issues and the possible underlying mechanisms remain severely limited.

In this study, CRSs are defined as long-lasting but potentially reversible transitions, reflecting persistent deviations from the climatic baseline rather than permanent shifts to a new equilibrium. Employing the sequential t-test analysis of regime shifts (hereafter STARS; see “Methods”)50,51, we evaluate and compare the CRS occurrence probabilities during super El Niño and non-super El Niño (hereafter non-SE) years for sea surface temperature (SST), surface air temperature (SAT), and surface soil moisture using multiple reanalysis datasets. We have three observed cases of super El Niño on record (i.e., the 1982/83, 1997/98 and 2015/16), and the results are verified by further analyzing the historical and future simulations of the Community Earth System Model version 2 Large Ensemble (CESM2-LE)52 and the Coupled Model Intercomparison Project Phase 6 (CMIP6)53 (Supplementary Table 1). We find that super El Niño significantly increases the likelihood of CRSs and may invoke more CRSs under global warming, suggesting important implications for the associated risk mitigation and climate adaptation strategies.

Results

Super El Niño-induced CRSs in observations

We begin by examining the baseline frequency of observed CRSs as detected by the STARS detection algorithm (see “Methods”). SST regime shifts are globally prevalent, with particularly high frequencies observed in extratropical oceans (Fig. 1a). By contrast, SAT regime shifts occur more often and are widespread over land, as well as ocean regions where synchronous SST shifts emerge (Fig. 1b). Surface soil moisture, which reflects the land water storage and its coupling to the climate system through soil moisture–atmosphere feedbacks54,55,56, also shows detectable regime shifts in several regions, with notable hotspots in the central Southern Asia, the western Amazon, East Africa, and central Australia (Fig. 1c). To further quantify this behavior, we estimated the long-term mean probabilities of regime shifts for each of the three climate variables (Supplementary Fig. 1a–c). These climatological annual probabilities are generally low, typically amounting to only a few percent, yet they capture the background likelihood of abrupt transitions under average conditions.

Fig. 1: Spatial patterns of climate regime shift (CRS) occurrence numbers and super El Niño (SE)-induced increases in CRS probability (∆prob) in observations.
figure 1

a–c Spatial patterns of total change-point numbers (shading) and super El Niño-induced change point numbers (blue contour; ≥ 2) for sea surface temperature (SST; ERSSTv5), surface air temperature (SAT; NCEP), and surface soil moisture (CPC), respectively. Change point numbers represent the total frequency of regime shift occurrences across four seasons: December–February (DJF), September–November (SON), June–August (JJA), and March–May (MAM). d–f Spatial patterns of increases in CRS probability during super El Niño events (∆prob), expressed relative to the climatological mean, for SAT, SST, and surface soil moisture, respectively.

Full size image

As ENSO acts as a key modulator of CRS occurrence, we further estimate the temporal CRS probabilities for seasonal and annual mean SST, SAT, and surface soil moisture (see “Methods”). During regular El Niño and La Niña events, probabilities show modest and spatially patchy increases relative to the climatological mean (Supplementary Fig. 1d–i), only marginally exceeding the climatological baseline. In addition, these signals remain inconsistent across these three climatic variables. This indicates that while regular ENSO events can moderately elevate the likelihood of CRSs, they are generally insufficient to trigger widespread and coherent CRSs.

In contrast, super El Niño events exert a far stronger and more spatially extensive influence on CRSs. The number of detected change points during these events is among the highest on record, with hotspots aligning with the regions characterized by a high frequency of climatological frequency of CRSs (Fig. 1a–c). To quantify this effect, we calculated the increase in regime shift probabilities during super El Niño years relative to the climatological mean (Fig. 1d–f; see “Methods”). Unlike regular ENSO events, which generate only relatively weak and localized signals (Supplementary Fig. 1), super El Niño years produce widespread and coherent enhancements in CRS probabilities across all three variables.

For SST, the most pronounced probability increases occur in regions strongly influenced by ENSO teleconnections43,57,58, including the central and western North Pacific, southeastern Indian Ocean, southwestern Pacific, Gulf of Mexico, and parts of the Atlantic (Fig. 1d). These basin-scale responses reveal extensive and intensified patterns of SST regime shifts, underscoring the unique capacity of super El Niño events to amplify abrupt oceanic transitions and reshape coupled atmosphere-ocean dynamics with far-reaching climate consequences.

In addition to the oceanic regions exhibiting synchronous SST regime shifts, super El Niño events also significantly elevate the probability of SAT regime shifts over land, particularly across ENSO-sensitive mid- and high-latitude regions such as East Africa, South America, and the Maritime Continent (Fig. 1e). These responses align with well-established teleconnection pathways through which ENSO modulates atmospheric circulation and heat fluxes59,60, and underscore the potential of super El Niño events to intensify extreme heat waves, with serious implications for ecosystems and human society61,62.

Similarly, super El Niño events markedly increase the probability of surface soil moisture regime shifts, with pronounced hotspots in ENSO-sensitive regions including the central southern Asia and central Australia (Fig. 1f). These increases are substantially stronger than those observed during regular ENSO events, underscoring the exceptional ability of super El Niño to disrupt terrestrial water balance. Given that many of these regions are already susceptible to aridification63,64, the amplification of abrupt soil moisture shifts by super El Niño further raises the risk of prolonged agricultural droughts.

As shown by the temporal evolution of the global mean CRS probabilities, besides the well-documented 1976–78 regime shifts in SAT and SST10,65, the three super El Niño events stand out with significantly higher values compared to non-SE years (Fig. 2). Although the specific distribution and magnitude of regime shifts exhibit seasonal variations, the overall increase in probability during super El Niño periods is robust and consistently detectable. This reinforces the conclusion that super El Niño events increase the likelihood of CRSs globally. It is important to note that the shifts were identified from which the long-term global warming trend has been removed. Nevertheless, across multiple datasets, the global CRS probability for surface soil moisture exhibits a distinct escalation trend over time, in addition to the sharp rise during super El Niño years. These results suggest that the occurrence probability of agricultural droughts may increase nonlinearly as the climate warms66, with super El Niño events acting as powerful amplifiers of these nonlinearities.

Fig. 2: Temporal evolution of annual climate regime shift (CRS) probabilities in observations and comparison between super El Niño and non-super El Niño years.
figure 2

Global mean CRS probabilities of seasonal- and annual-mean climatic variables: a Sea surface temperature (SST), b Surface air temperature (SAT) and c surface soil moisture. Super El Niño events are marked with blue boxes and red dots for the years 1982/83, 1997/98, and 2015/16. The color bars on the right represent the average CRS probabilities for non-super El Niño years (yellow) and during the three super El Niño events (cyan) for d SST, e SAT, and f surface soil moisture, respectively. DJF, MAM, JJA, SON, and annual represent December–February, March–May, June–August, September–November, and annual mean, respectively. Error bars indicate the 5th and 95th percentiles based on the bootstrap resampling method (see “Methods”). Note that the probabilities for the last five years have been set as missing values due to the limited data length to accurately detect regime shifts.

Full size image

We verify the robustness of these findings using multiple reanalysis datasets for all three climate elements (Supplementary Figs. 2–7; see “Methods”). The consistency of spatial patterns and temporal evolutions of CRS probabilities across different products suggests that the increased likelihood of CRSs during super El Niño events is a robust signal. We acknowledge that some uncertainty remains in surface soil moisture estimates, particularly in the ERA5-Land product (Supplementary Figs. 4 and 7), likely due to its dependence on ERA5 atmospheric forcing and the absence of direct soil moisture assimilation67.

As the detection of regime shifts may be sensitive to the cut-off length L and p-value used in the STARS method40, varying L (6, 8, 10, and 12) and p (0.01, 0.05, and 0.1) values are also used to test the robustness of super El Niño-CRS relationship (see “Methods”). Although different cut-off lengths affect the absolute level of detected CRS probabilities, the CRS consistently shows higher probabilities during super El Niño than during non-SE years. Similarly, while a stricter significance level (e.g., p = 0.01) results in fewer detected shifts due to a higher acceptance threshold, the conclusion that super El Niño significantly increases the likelihood of CRSs remains supported (Supplementary Figs. 8–12).

Evidence of super El Niño-driven CRSs and possible mechanisms

To further illustrate the CRSs associated with super El Niño events, we analyze the temporal evolution of key climate elements in several hotspot regions identified from the probability analysis (Fig. 1). In the central North Pacific, boreal winter SST underwent an evident regime shift following the 1997/98 super El Niño event, decreasing by approximately 0.8 °C in mean value (Fig. 3a). Similarly, SST anomalies over the western North Pacific transitioned abruptly from negative to positive after the 1997/98 and 2015/16 events (Fig. 3b). The southeastern Pacific also exhibited a shift to a negative SST phase after the 1982/83 and 1997/98 super El Niños (Fig. 3c). Additionally, SST over the Gulf of Mexico showed abrupt changes after the 2015/16 super El Niño event (Fig. 3d), which may have contributed to the recent intensification and increased frequency of hurricanes and storm surge hazards along Gulf Coast regions68,69.

Fig. 3: Evidence of super El Niño-induced climate regime shifts (CRSs) across various regions and climatic variables.
figure 3

Time series of standardized anomalies for sea surface temperature (SST), surface air temperature (SAT), and surface soil moisture (Soil) across selected regions with respect to 1948–2022 climatology. The vertical black lines and blue triangles denote the timing of CRSs in each variable associated with three super El Niño events significant at the 95% confidence level. The central map highlights the key regions selected as examples. a–d SST regions include the Central North Pacific (CNP), western North Pacific (WNP), southeastern Pacific (SEP) and Gulf of Mexico (GM); e–h SAT regions include the southwestern Pacific (SWP), Maritime Continent (MC), Eastern Africa (EAF), and northern Europe (NEU); i–l Surface soil moisture regions include the Central Australia (CAU), Central Asia (CAS), western Greenland (WGL) and the Amazon (AMA). “PAC” in the panel titles denotes “Pacific”. DJF, MAM, JJA, SON, and annual represent December–February, March–May, June–August, September–November, and annual mean, respectively. Pink and blue shaded areas denote positive and negative anomalies, respectively.

Full size image

Super El Niño events induce widespread SST regime shifts through multiple physical processes. In the central North Pacific, super El Niño initially triggers a strong cooling anomaly during boreal winter via atmospheric teleconnection57 (Supplementary Fig. 13a, b). This cooling persists through the “Reemergence” mechanism70,71,72 (Supplementary Text 1), wherein wintertime SST anomalies that disappear during summer recur in the following winter due to seasonal changes in the mixed layer. As the mixed layer deepens in the subsequent winter, subsurface temperature anomalies are re-entrained into the surface layer, sustaining negative SST anomalies, as observed after the 1997/98 and 2015/16 super El Niño events (Supplementary Fig. 13c, d). A similar process may also explain super El Niño-related SST regime shifts in the North Atlantic71,73.

Beyond SST-related changes, SAT also shows notable regime shifts. In ocean regions, these shifts reflect local SST forcing and air-sea heat exchange (Fig. 3e). Over land, significant SAT shifts occurred in the Maritime Continent following the 1997/98 and 2015/16 super El Niños (Fig. 3f). Summertime SAT over eastern Africa experienced abrupt increases after these events (Fig. 3g), likely due to consecutive La Niña conditions often following super El Niño events, which induce a persistent surface high pressure anomaly and extremely high temperatures via local positive feedback processes74. Boreal spring SAT over North Europe also experienced abrupt changes following the 1982/83 and 2015/16 super El Niño events, though the direction of change varied (Fig. 3h). Although SAT anomalies may fluctuate interannually in some regions across different events, the underlying shift in the climatological mean persists, reflecting a step-like transition in the baseline. These cases demonstrate that super El Niño can trigger both local and remote SAT regime shifts through direct thermal forcing and atmospheric teleconnections (Supplementary Text 2).

Surface soil moisture in central Australia shifted from drought to wet conditions after the 1997/98 super El Niño, with mean values changing from approximately −20 mm to 40 mm (Fig. 3i). Following the 2015/16 super El Niño event, the region experienced another shift from wet to dry conditions. In central Asia, the 1982/83 El Niño initiated a prolonged drought, potentially due to cumulative soil moisture effects from subsequent La Niña events75,76, while the 1997/98 El Niño event terminated this severe drought (Fig. 3j). Similarly, after the 2015/16 super El Niño, the western Greenland experienced a surface soil moisture deficit (Fig. 3k), likely attributable to atmospheric circulation changes and sea-ice loss slow-down77. The 2015/16 events also led to record-breaking temperatures and extreme drought in the eastern Amazon78, causing a prolonged soil moisture deficit through land-atmosphere coupling processes (Fig. 3l). These regional transitions highlight that super El Niño events can trigger surface soil moisture regime shifts by disrupting the local soil moisture balance through large-scale temperature or precipitation anomalies54,56,79,80,81. In humid regions, increased soil moisture enhances evaporation and latent heat flux, promoting rainfall through positive land-atmosphere feedback80,82. In arid zones, soil moisture anomalies affect atmospheric circulation and moisture convergence, influencing pluvial recovery. These processes prolong soil moisture memory and sustain anomalies even after ENSO phases have shifted (Supplementary Text 3 and Supplementary Figs. 14–16).

Despite the cyclical nature of ENSO, super El Niño events can trigger persistent climate shifts that reorganize regional baselines. Through nonlinear amplification and memory mechanisms—such as oceanic heat storage and land-atmosphere feedback—these events generate sustained anomalies that outlast the El Niño phase itself. Although the specific processes involved are complex and region-dependent, the robust and widespread evidence of post-super El Niño regime shifts highlights the need for further in-depth research to elucidate regional dynamics and underlying mechanisms.

Historical and projected super El Niño-induced CRSs in climate model simulations

Given the limited observational record of super El Niño events, we employ the CESM2-LE to conduct parallel analyses on the same climate variables (see “Methods”). Although CESM2-LE shows some differences in the spatial patterns of SAT and SST regime shift probabilities compared to the observations (Supplementary Fig. 17a, b)—likely due to the systematic model biases in representing ENSO teleconnections—super El Niño events significantly increase the CRS probabilities in several ENSO-sensitive regions such as the northern subtropical Pacific, the central North Pacific and the North Atlantic (Supplementary Fig. 17d, e). On a global mean basis, super El Niño increases the CRS probabilities by ~20% for both SAT and SST compared with non-SE years, a result significant at the 95% confidence level (Fig. 4a, b). Regime shifts in surface soil moisture are also robustly detected, with the spatial patterns in regions including Australia, North America, and parts of South America closely matching the observed shifts (Supplementary Fig. 17c, f). Super El Niño events significantly increase the global mean probabilities for surface soil moisture by 5–10% (Fig. 4c). Critically, these increases are substantially larger than those associated with regular ENSO events (Supplementary Fig. 18), underscoring the outsized influence of super El Niño in driving CRSs.

Fig. 4: Super El Niño events increase the probabilities of climate regime shift (CRS) in the Community Earth System Model version 2 Large Ensemble (CESM2-LE) under historical and Shared Socioeconomic Pathway 3–7.0 (SSP3–7.0) scenarios.
figure 4

a–c Comparison of CRS probabilities (%) between super El Niño and non-super El Niño events under the CESM2-LE historical (Present Day; blue) and SSP3–7.0 (Warm Climate; red) scenarios for a surface air temperature (SAT), b sea surface temperature (SST), and c surface soil moisture, respectively. The boxplots represent the CRS probabilities averaged across four seasons and annual values, with error bars indicating the 5th and 95th percentiles calculated using the bootstrap resampling method (see “Methods”).

Full size image

ENSO teleconnection and associated climate impacts are projected to intensify under global warming43,44,83,84,85,86, which implies that super El Niño-induced CRSs may also become more frequent and severe. To evaluate its future changes under greenhouse warming, the CRSs of these three variables are detected under SSP3–7.0 scenario in CESM2-LE. The spatial patterns of CRS probabilities indicate more frequent CRSs under global warming compared to the historical period (Fig. 4a–c and Supplementary Fig. 17g–i). The amplification is particularly pronounced for super El Niño events: although regular ENSO slightly increases CRS probabilities across some regions in a warmer climate, super El Niño events produce much stronger and more widespread increases, particularly for SAT and SST (Supplementary Fig. 19). As the climate impacts of super El Niño events are projected to strengthen further in the future45,48,49, both the frequency and spatial extent of associated CRSs are expected to increase. Results from CMIP6 models further support the causal relationship between super El Niño and CRSs identified in CESM2-LE (Supplementary Fig. 20).

We note that under warming scenarios, non-SE years can trigger surface soil moisture regime shifts at levels comparable to those during super El Niño events (Fig. 4c). This suggests that a warmer climate renders the land soil system more sensitive, amplifying land–atmosphere feedbacks56,80. However, substantial uncertainties remain, due in part to limitations in simulating soil moisture-atmosphere interactions in climate models, which are compounded by a severe scarcity of in situ soil moisture observations across many regions66,87.

We hypothesize that the robust increase in super El Niño-induced CRSs under global warming arises from the intensification of ENSO climatic impacts. For example, in the central North Pacific, the ENSO-related SST response is projected to strengthen, possibly due to shallower mixed layer with reduced heat capacity88. Through the “Reemergence” mechanism71,72,89, SST anomalies can persist longer by being stored in the subsurface during summer and reemerging in winter, thereby increasing the likelihood of regime shifts on multi-year to decadal timescales. The increase in SAT regime shifts is closely tied to the synchronous SST transitions, mediated through both local ocean-atmosphere coupling and remote teleconnections. For surface soil moisture, intensified temperature-soil moisture feedbacks may prolong anomalies by amplifying low-frequency SAT variability81. In addition, enhanced super El Niño-driven precipitation in a warmer climate may reinforce soil moisture-precipitation feedbacks44, further driving abrupt soil moisture regime shifts. The detailed mechanisms underlying these responses warrant further investigation.

Discussion

Using multiple observational datasets and climate model simulations, this study reports that super El Niño events significantly increase the likelihood of CRSs in terms of SST, SAT, and surface soil moisture. These shifts are likely driven by the ability of super El Niño to promote extreme climate and weather events, with critical implications for global climate variability, hydrological cycles, and ecosystem stability. As the climatic impacts of super El Niño events are projected to intensify under global warming45,48,49, the resulting regime shifts in these key variables are expected to become more widespread and pronounced.

We further propose that super El Niño may act as a catalyst for transitions in inherent modes of decadal climate variability90,91. For example, super El Niño can generate persistent SST anomalies over the Central North Pacific through the “Reemergence” mechanism71,72,89, potentially initiating phase transitions of the Pacific decadal oscillation (PDO). Notably, the observed negative phase shifts of the PDO in the late 1990s and in 2016 align closely with the 1997/98 and 2015/16 super El Niño events. Given that future projections suggest declining predictability of decadal variability (e.g., the PDO) under global warming92, the precursory role of super El Niño in triggering such transitions demands greater attention. A detailed investigation into the specific mechanisms through which super El Niño induces CRSs is essential, as these processes are likely case-dependent and region-specific.

Although our analysis concentrates on physical climate variables, super El Niño events may also provoke persistent or even irreversible changes in other components of the climate and Earth system, including sea ice dynamics and marine and terrestrial ecosystems. For example, ENSO has been shown to significantly influence Arctic sea ice thickness and concentration93, as well as Antarctic shelf ocean warming and sea ice melting94. Against the backdrop of pronounced Arctic warming—which has accelerated sea ice melting and caused a regime shift in the Arctic sea ice thickness95—super El Niño may further induce shifts in sea ice concentration and extent across both hemispheres (Supplementary Fig. 21), thereby amplifying warming and melting through positive feedbacks96,97,98. In extreme cases, super El Niño may trigger abrupt transition from a quasi-stable ice shelf front to catastrophic large-scale disintegration due to regional sea ice loss99, contributing to sea-level rise. A key challenge for future research is to determine the mechanisms that govern whether super El Niño-induced transitions remain persistently reversible or cross critical thresholds into irreversible change, which carries profound implications for decadal predictability and societal adaptation.

CRSs have profoundly affected climate systems, ecosystems, human livelihoods, and the global economies. As global warming is expected to strengthen the impacts of super El Niño events, the probability and magnitude of CRSs are projected to rise nonlinearly, thereby heightening vulnerabilities across environmental and socioeconomic systems. By identifying regions and climate variables most susceptible to these shifts, our findings provide critical insights for the development of advanced early warning systems, which are indispensable for proactive and effective climate response. Furthermore, these results can inform targeted adaptation strategies to strengthen the resilience of vulnerable ecosystems and communities, helping to mitigate the adverse impacts of abrupt climate transitions.

Methods

Observations

We used multiple reanalysis products of sea surface temperature (SST) and surface air temperature (SAT) to detect the regime shifts in observations from 1948 to the present. The SST reanalysis datasets are from: (1) Hadley Centre Sea Surface Temperature dataset version 1.1 (HadISSTv1.1)100; (2) National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed Sea Surface Temperature version 5 (ERSSTv5)101; (3) Centennial in situ Observation-Based Estimates Sea Surface Temperature version 2 (COBE-SST v2)102; (4) Kaplan Extended Sea Surface Temperature version 2 (Kaplan SST v2)103; (5) the merged Hadley Centre-NOAA-Optimal Interpolation (version 2; OI.v2) sea surface temperature (SST) and sea ice concentration (SIC) datasets (HURRELL SST)104; (6) European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis version 5 SST (ERA5 SST)105; (7) The National Centers for Environmental Predictions/Department of Energy (NCEP/DOE) AMIP-II Reanalysis (Reanalysis-2) SST (AMIPII)106.

The SAT datasets are derived from: (1) the National Centers for Environmental Prediction (NCEP) Reanalysis107; (2) The NOAA Global Surface Temperature Dataset (NOAAGlobalTemp) which is a merged land-ocean surface temperature analysis (MLOST)108; (3) the ECMWF Reanalysis v5 (ERA5)105; (4) Goddard Institute for Space Studies Surface Temperature Analysis version 4 (GISTEMP v4)109; (5) Hadley Centre/Climatic Research Unit, version 4 (HadCRUT4)110; (6) Global land surface temperatures combining two large individual datasets of station observations collected from the Global Historical Climatology Network version 2 (GHCN) and the Climate Anomaly Monitoring System (CAMS) (refer to as GHCNCAMS)111.

We used a suite of observation-based and reanalysis surface soil moisture products to detect regime shifts from 1948 to 2023. The datasets include: (1) NOAA Climate Prediction Center (CPC)112 (1948–2023); (2) NCEP/NCAR reanalysis107 (1948–2023); (3) TERRAclimate113, a high-resolution (~4 km, 1/24°) monthly climate and water balance dataset of global land surface (1958–2023); (4) ERA5105 (1948–2023); (5) ERA5-land, a high-resolution land surface reanalysis dataset produced by replaying the land component of ERA5 using its atmospheric forcing (1950–2023); (6) The European Space Agency’s (ESA) Climate Change Initiative (CCI) Soil Moisture COMBINED product merging active and passive satellite retrievals (ESACCI; 1978–2023)114; (7) the Global Land Evaporation Amsterdam Model (GLEAM; 1980–2023)115; (8) NASA Global Land Data Assimilation System Version 2 (GLDAS; 1948–2014)116; (9) the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2; 1980–2023)117; (10) self-calibrating Palmer drought severity index (SCPDSI) from Climatic Research Unit (CRU) (1948–2023)118. Monthly precipitation data from the NCEP reanalysis107 is incorporated to understand surface soil moisture regime shifts.

Sea ice data sourced from the Met Office Hadley Centre and the National Snow and Ice Data Center (NSIDC) were applied for sea ice regime shift detection. The Hadley Centre provides monthly sea ice concentration, and the NSIDC includes sea ice extent, covering the period 1979–2023.

The Pacific Decadal Oscillation (PDO) time series from National Centers for Environmental Information (website: https://www.ncei.noaa.gov/access/monitoring/pdo/) based on NOAA’s ERSSTv5 was utilized to explore the potential role of super El Niño events in driving decadal phase transitions.

Climate models

In this study, we utilized the Community Earth System Model version 2 Large Ensemble (CESM2-LE)52,119 to validate observational findings and assess future projections in regime shifts. CESM2 is a fully coupled Earth system model that incorporates the CAM6 atmosphere model and POP2 ocean model, both on ~1° horizontal grids, as well as coupled land, sea ice, wave, marine biogeochemical, and river runoff models. The CESM2-LE consists of 100 ensemble members run from 1850 to 2100 and forced under the Coupled Model Intercomparison Project Phase 6 (CMIP6) historical protocols (1850–2014) and Shared Socioeconomic Pathway 3–7.0 (SSP3–7.0) scenario (2015–2100). To isolate internal climate variability, anomalies for each variable were computed by removing the ensemble mean from each member and applying quadratic detrending.

Additionally, we examined regime shifts in SST, SAT, and surface soil moisture across 48 members of CMIP6 from various Coupled Global Climate Models53 (CGCMs; Supplementary Table 1). We compared simulations under historical forcing (1850–2014) and the high-emission SSP5–8.5 scenario (2015–2100) to evaluate changes in the probability of regime shifts associated with super El Niño events under global warming. For each model and period, the climatological mean was removed, and quadratic detrending was applied for regime shift detection.

Regime shift detection

The sequential t-test analysis of regime shifts (STARS) was systematically applied to the full time series of each variable at every grid point across all observational and model-based datasets used in this study8,12,50,51,95,120,121. This method enables the identification of climate regime shifts (CRSs) in a purely data-driven manner, without requiring prior assumptions regarding the timing or number of shifts.

In the STARS algorithm, a regime is defined as the sequence of consecutive data points since the most recent detected shift (or the start of the record). For each subsequent observation, its anomaly relative to the current regime mean is evaluated using a Student’s t-test to determine whether it differs significantly from that mean. If a significant deviation is detected, subsequent data points are examined sequentially to assess the persistence of the change. The regime shift index (RSI) is computed as the cumulative sum of normalized anomalies from the hypothetical mean of the new regime. A shift is confirmed when the RSI remains non-zero for at least the specified cut-off length (L), indicating a sustained deviation from the prior mean state.

For our primary analysis, the cut-off length L was set to 10 years to exclude short-term excursions—such as those associated with interannual ENSO variability (typically lasting 1–2 years)—that do not meet the persistence criterion required for a regime shift. It is important to note that while the temporal behavior following a shift may vary across variables and regions (e.g., manifesting as a stable new mean or oscillations around that mean), the defining feature of all detected regime shifts is a statistically significant and persistent change in the long-term mean state, irrespective of subsequent short-term variability. In this framework, CRSs identified through STARS are defined operationally as statistically significant and persistent (≥ L years) deviations from the climatic baseline. They do not represent irreversible tipping points, but rather long-lasting anomalies that may be sustained yet remain potentially reversible.

To evaluate the sensitivity of our results, we conducted additional analyses using alternative cut-off lengths (L = 6, 8, and 12 years) and varying statistical significance levels (p = 0.01, 0.05, 0.1). Years in which RSI values remain non-zero for the entire cut-off length were classified as regime shift years. Further details regarding the STARS methodology and its application in climate studies can be found in previous publications8,13,50,51,95,120,121,122.

Regime shift occurrence probability calculation

To enhance the statistical robustness of our analysis, we detected the CRSs in both observational and climate model datasets for the annual mean and for the four meteorological seasons: December–February (DJF); March–May (MAM); June–August (JJA), and September–November (SON). For each grid cell, the total probability of a CRS occurrence \({{{{\rm{Prob}}}}}_{t}\left(x,y\right)\) was calculated as:

$${{{{\rm{Prob}}}}}_{t}\left(x,y\right)=\frac{n\left(x,y\right)}{T}$$
(1)

where \(n(x,y)\) is the number of years in which a regime shift was detected at grid location \((x,y)\), and \(T\) is the number of total years in the time series. To specifically quantify the influence of super El Niño events, we defined the CRS probability due to super El Niño events (\({{{{\rm{Prob}}}}}_{{SE}}\)) by dividing the occurrence numbers of super El Niño-induced regime shifts \({n}_{{SE}}\) by the number of total super El Niño years \({T}_{{SE}}\). We attributed a regime shift to a super El Niño event if it occurs during either the development year or the subsequent year of a super El Niño event. In this case, in observations, there are six super El Niño years (1982–83, 1997–98, and 2015–16). Thus, the increased probability due to super El Niño at each grid for each season was then computed as:

$$\triangle {{{\rm{Prob}}}}\left(x,y\right)={{{{\rm{Prob}}}}}_{{SE}}-{{{{\rm{Prob}}}}}_{t}=\frac{{n}_{{SE}}\left(x,y\right)}{{T}_{{SE}}}-\frac{n\left(x,y\right)}{T}$$
(2)

This approach allows for precise mapping of the spatial and seasonal patterns of super El Niño-induced regime shifts. In CESM2-LE, results were aggregated across all 100 ensemble members to compute probabilities, significantly improving the robustness and representativeness of the estimates.

To obtain an integrated global perspective, the annual CRS probability was estimated by determining the total regime shift numbers at each latitude \(\varphi\) by considering the latitude weights. The CRS probability in the i-th year is given by:

$${{{\rm{Prob}}}}\left(i\right)=\frac{{\int }_{-\frac{\pi }{2}}^{\frac{\pi }{2}}\cos \varphi \cdot n(i,\varphi )d\varphi }{{\int }_{-\frac{\pi }{2}}^{\frac{\pi }{2}}\cos \varphi \cdot N(\varphi )d\varphi }$$
(3)

where \(n(i,\varphi )\) is the number of CRSs detected at each latitude \(\varphi\) during the i-th year, and \(N(\varphi )\) represents the number of grid cells at each latitude \(\varphi\). This latitude-weighting ensures an accurate representation of the temporal evolution of regime shifts by accounting for the area distribution across latitudes.

Definition of super El Niño events

The Niño3 index, defined as the area-averaged sea surface temperature anomaly (SSTA) over the Niño3 region (5°S–5°N, 150°W–90°W), is used to identify the super El Niño events that are characterized by strong variability in the equatorial eastern Pacific. The SSTA was calculated by removing the quadratic trend and the monthly climatological mean for each reanalysis dataset and each ensemble in CESM2-LE and CMIP6. A super El Niño event is identified when the normalized Niño3 SSTA during the ENSO mature phase (December-February, DJF) exceeds a threshold of 2-standard deviation.

Statistical significance test

We used a bootstrap method to examine whether the regime shifts due to the super El Niño events are statistically significant123. (1) For observations, we estimated the CRS probability and super El Niño-induced increase in probability for SST, SAT, and surface soil moisture by randomly sampling a number of data points equal to the count of regime shifts with replacement 10,000 times. (2) For model ensembles (CESM2-LE and CMIP6), we first aggregated regime shift results across all ensemble members. We then generated 20,000 synthetic realizations via random resampling with replacement across the full analysis period under both historical and future forcing scenarios (SSP3–7.0 and SSP5–8.5). This approach allows us to evaluate the significance of super El Niño-related CRSs under both current and projected climate conditions.