The Antarctic Circumpolar Wave (ACW) is the dominant interannual signal in the Southern Ocean, influencing global climate, yet its spatiotemporal structure and physical mechanisms remain debated. Historically, the ACW was described as an eastward-traveling wave with a period of 4–5 years and wavenumber 2, but different analysis methods have yielded contradictory results: some studies found a standing dipole, others a traveling wave, and still others a combination of signals. These discrepancies partly stem from subjective data filtering, which can mix signals from different physical processes. This work aligns with research that applies advanced climate data analysis methods to separate tropical influences from local ocean–atmosphere interactions, similar to studies on gravitational waves or neutron stars, where extracting weak coherent signals from noise is also crucial.
The authors applied nonlinear Laplacian spectral analysis (NLSA), a method grounded in dynamical systems theory and kernel-based machine learning. Unlike classical EOF analysis, NLSA operates on delay-embedded data and uses anisotropic cone kernels that account for local dynamics. This allows simultaneous extraction of multiple time scales without preprocessing: no trend removal, no deseasonalization, and no bandpass filtering. They analyzed sea surface temperature (SST) and sea ice concentration (SIC) data from a 1300-year control run of the CCSM4 model and HadISST observations spanning 1979–2017. To investigate tropical teleconnections, SST data from the Indo-Pacific region were analyzed separately. Additionally, 850-hPa geopotential height anomalies were reconstructed to identify Rossby wave trains.
NLSA identified a hierarchy of modes: annual cycle modes, interannual ACW modes and their combination modes, and decadal and trend modes. The primary ACW mode with wavenumber 2 has a peak frequency near 0.25 cycles per year (period ~4 years) and propagates eastward. At certain phases, its spatial structure forms a dipole between the eastern Pacific and the western Atlantic, explaining the statistical origin of the Antarctic Dipole, previously thought to be a standing wave. A family of combination modes was found with frequencies of the form |m·$f_{ACW}$ + n·$f_A$|, where $f_A$ is the annual cycle. These modes can propagate both eastward and westward and exhibit a meridional pulsation reminiscent of Arctic sea-ice reemergence phenomena. They explain more variance in the Indian Ocean than the primary ACW modes. The correlation between the wavenumber-2 ACW mode and ENSO modes extracted from tropical data reaches 0.9, confirming teleconnection via the Pacific–South American wave train. The wavenumber-3 wave showed no significant correlation with the tropics, indicating a local origin.
This work demonstrates that contradictions in previous ACW studies are largely explained by subjective data filtering. The objective NLSA method, without preprocessing, reveals that the Antarctic Dipole is not a separate mode but a phase of a traveling wave. This changes our view of Southern Ocean dynamics: instead of a static dipole, there is a continuously propagating wave with variable amplitude. Furthermore, for the first time, a complete hierarchy of combination modes with the annual cycle has been extracted from Antarctic data, opening the door to studying nonlinear interactions in the ocean–atmosphere–ice system. Unlike neighboring studies, such as those on gravitational waves or neutron stars, this work applies NLSA, a method specifically designed for climate data that can handle multiscale variability without a priori assumptions.
Plans include refining and further validating the identified patterns using recently developed Koopman operator methods for high-dimensional spatiotemporal climate data. The close correlation between ACW and ENSO modes also holds promise for predictive studies, for instance, by extending NLSA for analog forecasting. This could improve seasonal and interannual forecasts of Southern Ocean and sea-ice conditions.
The results will influence our understanding of global climate variability, particularly in the Southern Hemisphere, and the interpretation of observational and model data. They may also aid in improving parameterization schemes in climate models that account for nonlinear interactions between the annual cycle and interannual modes.
The next step is to apply Koopman operator methods to the same data to test the reproducibility of the mode hierarchy. Additionally, the development of a predictive model based on the identified modes is planned for forecasting temperature and sea-ice anomalies in the Southern Ocean.
This work directly addresses an unresolved problem in Southern Ocean climatology: separating the contributions of tropical forcing (ENSO teleconnections) and local ocean–atmosphere interactions in shaping the Antarctic Circumpolar Wave. It also tackles the issue of objectively extracting coherent structures from noisy climate data, which is important across many areas of climate science.
The Antarctic Circumpolar Wave circles Antarctica in about 8 years, carrying temperature and ice anomalies at speeds of about 6–8 cm/s, comparable to a walking pace.