Largest Eigenvalue of Ocean SST Correlation Matrices Rises in Lockstep with Global Mean Temperature Since 1980
Spectral analysis of annual SST correlation matrices reveals a secular rise in the leading eigenvalue that directly indexes strengthening collective ocean-temperature correlations. The Gaussian rather than Tracy-Widom statistics of this eigenvalue indicate the system has moved outside the weakly correlated regime assumed by conventional climate diagnostics. These hidden covariance patterns provide a sensitive complement to mean-temperature metrics for detecting long-term climate change.
The arXiv preprint constructs 12-month correlation matrices from daily SST fields across 1° grids and applies random-matrix diagnostics to each annual ensemble. Eigenvalue densities deviate systematically from the Marchenko-Pastur bulk while the largest eigenvalue follows a near-Gaussian distribution rather than Tracy-Widom statistics, indicating strong collective modes driven by basin-scale warming. This spectral shift captures the progressive strengthening of long-range temperature correlations that standard linear trend maps obscure.
Standard reporting of global mean SST rise misses the reorganization of spatial covariance structure. The paper’s finding that the leading eigenvalue tracks cumulative ocean heat content more closely than interannual variability aligns with independent analyses in Nature Climate Change (2023) showing emergent spatial coherence in marine heatwave footprints and with earlier random-matrix work on atmospheric pressure fields that likewise detected non-universal spectral tails under greenhouse forcing.
Because the largest eigenvalue is sensitive to both amplitude and spatial extent of anomalies, it may serve as an early-warning observable for regime shifts in ocean circulation. Continued monitoring through 2035 will test whether the observed linear rise saturates once warming exceeds 1.5 °C above pre-industrial levels.
Next steps require extending the ensemble to subsurface temperature profiles and testing whether the same spectral inflation appears in CMIP6 historical runs under different forcing scenarios.
Da Silva: largest eigenvalue exceeds 55 by 2032 if annual SST rise remains above 0.12 °C decade⁻¹
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2606.14916)
- [2]Supporting Source(https://www.nature.com/articles/s41558-023-01685-4)