The impacts of air-sea and convective coupling on the daily predictability of tropical atmospheric anomalies

The effect of air-sea and convective (moisture) coupling on daily tropical climate variability is investigated in a coupled linear inverse model (LIM) derived from the simultaneous and 1-day lag covariances of observed ERA-Interim daily average departures of winds, heights, specific humidity, sea level pressure, and SST from the annual cycle. The model predicts the covariances at all other lags, as well as OLR covariances from a regression of olr on the model state variables. The predicted and observed lag covariances, as well as the associated power spectra and Wheeler-Kiladis diagrams, are generally found to agree within sampling uncertainty. This validation justifies a linear diagnosis of predictability and coupling in the system.

The results show that air-sea coupling generally has a very small effect on daily-to-subseasonal tropical atmospheric predictability ranging from Kelvin waves to the MJO. It has much larger effects on longer-term variability, in both the atmosphere and ocean, including greatly increasing the amplitude of ENSO and lengthening its dominant period from two to four years. Consistent with these results, the eigenvectors of the system's dynamical evolution operator also separate into two distinct, but nonorthogonal, subspaces: one governing the nearly uncoupled daily-to-subseasonal dynamics, and the other governing the strongly coupled longer term dynamics. These subspaces arise naturally from the LIM analysis; no bandpass frequency filtering is applied. One implication of this remarkably clean separation of the uncoupled and coupled dynamics is that GCM errors in anomalous tropical air-sea coupling may cause substantial errors on interannual and longer time scales, but probably not on much shorter time scales such as those associated with the MJO. The moisture coupling is not so cleanly separated into different spaces, but its effects are clearly demonstrated by comparing the eigenmodes of two dynamical systems: the original LIM and a second system in which coupling of all other atmospheric variables to specific humidity anomalies is removed. Finally, the most predictable tropical states are determined, both for the full LIM and the uncoupled LIMs, and compared to the realized forecast skill.