The climate community has long debated how much natural variability, particularly El Niño, can explain observed warming. Previous studies, such as McLean et al. (2009), found high correlations between the Southern Oscillation Index (SOI) and tropospheric temperature, but because they removed trends, they could not link ENSO to long-term warming. Foster et al. (2009) criticized this approach, arguing that SOI explains only 15–30% of interannual variability and has little effect on the trend. In this work, the authors introduce a new variable, the cumulative SOI (cSOI), which accumulates the effects of El Niño and La Niña over time. They rely on physical mechanisms such as the asymmetry between El Niño and La Niña, positive feedbacks through clouds and winds, and statistical evidence of random walks in temperature series. The work stands apart from neighboring studies on the solar corona and solar wind but resonates with work on El Niño prediction and precipitation regime shifts, expanding our understanding of the ocean's role in the climate system.
The authors used multiple linear regression to model global temperature. As dependent variables, they took two datasets: RATPAC-A (tropospheric temperature, smoothed with a 12-month filter, as in McLean et al.) and HadCRUT3 (surface temperature without filtering). Independent variables included the SOI itself, its cumulative sum cSOI (the sum of all previous SOI values since the start of the series), sunspot number (SSN) as a proxy for solar radiation, stratospheric aerosols (V), and anthropogenic forcing (AF), which is a smoothed sum of all factors except volcanism and solar activity. The SOI was not time-shifted (no lag). To test model robustness, they applied cross-validation by splitting the sample into two parts at 1984: coefficients of determination R² were calculated both within the sample and on the held-out portion. Regression coefficients allowed estimation of the SOI accumulation fraction via the ratio of coefficients for cSOI and SOI.
A model with only SOI explains just 7% of tropospheric temperature variation (RATPAC-A) from 1960 to 2008. Adding the cumulative cSOI raises the explained variance to 50%. A model with only the anthropogenic factor (AF) explains 69%, and adding cSOI and SOI increases it to 71%. The full model with all variables explains 78% of the variation. For surface temperature HadCRUT3, cSOI explains 54% of the variation, and adding SOI to the anthropogenic model improves it from 66% to 72%. The full model explains 77%. Cross-validation yields lower but consistent R² values, confirming robustness. The estimated SOI accumulation fraction from the cSOI/SOI coefficient ratio is 4.9±1.5% (when AF is included) and up to 9±2% (without AF). The cSOI range from 948 in 1976 to -172 in 1998 corresponds to a global temperature change of 0.8817°C, giving a sensitivity of 0.00079°C per SOI unit. The sensitivity to the anthropogenic factor when cSOI is included is estimated at 0.19±0.02 K/(W/m²), which is lower than the 0.30 K/(W/m²) for the zero-inertia response to CO₂, but significantly lower than the 0.8 K/(W/m²) required by models to reproduce the trend.
The work shows that the accumulated effect of El Niño can explain up to half of the observed warming since 1960, challenging the unequivocal dominance of anthropogenic greenhouse gases. Unlike neighboring studies that focus on short-term correlations or individual mechanisms, this work introduces for the first time the cumulative variable cSOI, which captures multi-decadal trends. This changes the perspective: natural internal climate variability may be a much more powerful factor than previously thought. The results are consistent with observations of reduced cloudiness during the positive phase of the Pacific Decadal Oscillation and with the hypothesis of a random walk in temperature that integrates oceanic anomalies. If the estimate is correct, climate models that do not reproduce such accumulation may systematically overestimate the anthropogenic contribution.
The topic could develop toward more detailed modeling of heat accumulation mechanisms in the ocean, especially considering the reappearance of anomalies in the following winter season (the Alexander and Deser mechanism). Models with a multi-layer ocean capable of reproducing anomaly transport through mixing will be needed. It is also important to investigate the interaction between ENSO and anthropogenic forcing: if they are not independent, interpretation becomes more complex. Long-term paleoclimate reconstructions of cSOI could test whether such accumulations occurred in the past without anthropogenic influence. Finally, longer data records and expanded spatial coverage are needed to separate natural variability from the forced signal.
The results affect climate science, particularly the attribution of warming and the calibration of climate models. They are also important for policy, as they question the degree of confidence in anthropogenic dominance, which could influence adaptation and mitigation strategies.
Conduct an analysis with longer time series, including paleodata, and with spatially distributed temperature fields. Develop physically based models of SOI accumulation that account for ocean dynamics and feedbacks.
The work is directly related to the unresolved problem of global warming attribution: how to separate the contributions of natural internal variability and anthropogenic forcing when variables are collinear. It also raises the question of why climate models may underestimate decadal variability associated with oceanic integration of anomalies.
The cumulative SOI index peaked at 948 in 1976, coinciding with one of the coldest periods in the troposphere over the past 50 years, while the minimum of -172 in 1998 coincided with the record warmest year.