Optimal forcing of ENSO either side of the 1970’s climate shift and its implications for predictability

Inverse methods are used to investigate changes in the precursors to El Niño Southern Oscillation (ENSO) events since the so-called 1970’s climate shift, associated with a change in the phase of the Interdecadal Pacific Oscillation (IPO). Linear Inverse Models (LIMs) constructed from tropical sea surface temperature, thermocline depth and zonal wind stress anomalies from each of the periods 1959–1978 and 1979–1998, are able to reproduce the major observed characteristics of ENSO, including its amplitude, frequency and time evolution. Each LIM possesses low-frequency and biennial ENSO modes, the former being both the least damped and the mode responsible for strongest pseudoresonance, as quantified via calculation of the resolvent norm. Because these modes are damped, ENSO variability is sustained in the stochastically forced LIMs by transiently growing perturbations, and predictability is determined by the character of the transiently growing subspace of perturbations. The optimal linear precursor over any given lead time is equivalent to the optimal perturbation of the LIM, that represents the most rapidly growing linear perturbation over that timescale. In both periods linear ENSO growth occurs through one of two trajectories associated with the 7 and 15 month optimal perturbations. The structure of these two optimal perturbations change significantly between the two periods, and their ability to predict ENSO degrades dramatically when applied to the alternate period. This suggests that ENSO precursors changed following the 1970s climate shift over both 7 and 15 month time-scales. In particular, while prior to the climate shift the heat content of the equatorial Pacific alone is a skillful ENSO predictor on 7 month lead times, afterwards Indian and south Atlantic sea surface temperature anomalies are inferred to have become important. Optimal ENSO growth over 15 months also contains a significant extra-Pacific contribution, and it is possible to skillfully hindcast some (but not all) ENSO events in both periods over 15 month lead times. Of the four considered linear precursors, only the 7 month optimal perturbation corresponding to the period 1959–1978 is able to skillfully hindcast ENSO amplitude from 1998 to the present, correctly predicting the development of El Niño conditions since February 2014. As such the optimal precursor structure appears to be related to the phase of the IPO, and we conjecture that extra-Pacific teleconnections may gain importance during a positive phase of the IPO