Time-Lagged Climate Teleconnections
Context
The climate system resembles a conversation where the other party replies not immediately but after a pause: the ocean warms up, and only months later does the rainfall pattern change on another continent. Classical methods for finding links between data fields, such as maximum covariance analysis, are tuned to instantaneous correlation and miss these delays. There are already studies on the ocean conveyor belt and early climate signals, but they do not solve the problem of simultaneously extracting spatial patterns and temporal shifts. Our method stands apart: it combines the idea of complex numbers to account for phase shifts with rotation for physical interpretation, which neighboring approaches lack.
Methods
First, time series of sea surface temperature and land precipitation were smoothed with a six-month moving average and normalized by the standard deviation at each point. Then each series was turned into an analytic signal using the Hilbert transform, which adds an imaginary part shifted by a quarter period. To avoid edge artifacts, the series were extended into the past and future using the Theta model (ARIMA with drift), and after the transform they were trimmed. Next, the covariance matrix between the two complex fields was computed and singular value decomposition was applied. The first 150 modes, explaining 99.82% of the covariance, were subjected to Varimax rotation to localize patterns. The method is implemented in the xmca package in Python.
Results
The method identified six main modes of coupled ocean–precipitation variability. The first mode (62.7% of covariance) describes the seasonal cycle: the phase function allowed calculating that the seasonal temperature maximum in the northern Pacific lags the southern one by 196 days, and in the Indian Ocean precipitation lags temperature by 81 days. The second mode (7.0%) corresponds to El Niño: correlation with the ONI index was 0.941. The third mode (6.9%) reflects global warming with a correlation of 0.781 with the ocean warming index. The fourth (1.6%) reproduces the Pacific Decadal Oscillation (correlation 0.565), the fifth (1.6%) El Niño Modoki (0.506), and the sixth (1.1%) the Atlantic Meridional Mode (0.741). For modes with broadband spectra, the phase is interpreted only as correlation or anticorrelation, but for the seasonal cycle with a narrow peak at 12 months, the phase directly translates into a delay in days.
Implications
Unlike neighboring studies that either seek instantaneous links or fix the delay in advance, this method simultaneously extracts spatial patterns and their relative temporal shifts from a pair of fields. It is the first to combine complex signal representation with Varimax rotation within maximum covariance analysis. This allows one procedure to obtain all major teleconnections that previously were found by disparate regional studies. The method does not require prior detrending, which is critical in a changing climate.
Future development
The method opens the way to systematic study of seasonal and subseasonal phenomena, such as the Madden–Julian oscillation, where convection waves propagate eastward with phase shifts between variables. Plans include applying complex rotated MCA to less studied variable pairs: sea surface salinity, sea surface height, soil moisture, and wind. This may reveal new teleconnections and refine the dynamics of known modes. The xmca package makes the method accessible to a broad range of researchers.
Impact
The method will be useful in climatology, oceanography, and hydrology for finding lagged relationships between any geophysical fields. It may also improve statistical rainfall forecasting models that rely on slow oceanic predictors.
Next steps
The immediate next step is to apply the method to the Madden–Julian oscillation and check how well the phase function reflects the actual propagation of convection. Then, the sensitivity to the choice of the Theta model period and the number of rotated modes should be investigated.
Key open problems
An open question in climatology remains the physical mechanisms of teleconnections: which are caused by the atmospheric bridge and which by oceanic waves. The method provides a tool for testing hypotheses about delays, but it does not answer the question of causality. Another unsolved problem is separating natural variability from the anthropogenic trend in broadband modes.
The seasonal temperature maximum in the northern Pacific occurs 196 days later than in the southern Pacific, which is almost exactly half a year. This means the two ocean hemispheres work in antiphase, like a seesaw with a very slow period.
The idea of delayed teleconnections recalls the 'butterfly effect' from Ray Bradbury's story 'A Sound of Thunder': a small event in one place, through a chain of connections, leads to major consequences elsewhere. Only here the connections are not chaotic but statistically stable.
Key numbers
- Pacific seasonal maximum lag: 196 days
- El Niño mode correlation with ONI: 0.941
- Warming mode correlation with OWI: 0.781
- Explained covariance of the first 72 modes: >99%
- Number of rotated modes: 150