Assessing the Impact of Stochastic Forcing on ENSO Events

Andrew T. Wittenberg
Atmospheric & Oceanic Sciences Program, Princeton University

Eli Galanti
International Research Institute for Climate Prediction, Columbia University

2004 Joint Assembly, Montreal, Canada, 17-21 May 2004

Abstract: Previous work suggests that ENSO may be influenced by a highly chaotic, essentially stochastic, component of atmospheric variability.  To assess the possible impact of this "noise" on ENSO evolution and forecasts, we perform a suite of ENSO hindcasts spanning the past two decades.  The hindcast model consists of an ocean general circulation model (GFDL MOM4), coupled to a statistical atmosphere derived by regressing historical air-sea fluxes onto historical sea surface temperature anomalies (SSTAs).  The residual fluxes not captured by the statistical atmosphere are linearly independent of large-scale SSTAs, providing a rough estimate of the historical stochastic forcing.  A simple ocean data assimilation provides initial conditions for a suite of year-long stochastic ensemble hindcasts.  In a given hindcast, ensemble members begin from identical initial states but are forced by residual fluxes taken from different years.  One particular member, a "cheatcast," is forced by the actual residual observed during that year, and an additional "noise-free" hindcast is performed with the residual fluxes turned off.

The results show that both the initial conditions and the residual stress forcing affect the subsequent evolution of ENSO.  The initial state preconditions the tropical Pacific toward development of ensemble-mean SSTAs, while the residual forcing induces a rapid dispersion of the hindcasts about their ensemble mean.  To some extent, the residual fluxes induce similar effects regardless of the ocean preconditioning; in particular, the ensemble member forced by the 1997 residual is always among the warmest hindcasts, while that forced by the 1988 residual is always among the coldest.  Initializing from different years alters the dispersion of the ensemble, indicating that some oceanic nonlinearity is present.  However, the noise-free hindcast in all cases closely tracks the stochastic ensemble mean, suggesting that the hybrid coupled system is reasonably linear.  To clarify the role of nonlinearity in the wind stress response to SST, the statistical atmosphere is fit to an ensemble of atmospheric GCM simulations forced by observed SSTs.  The results indicate that in fact the residual stress is not all noise, but partly a teleconnection from outside the tropical Pacific or a nonlinear response to SST.  Implications of the results and limitations of the methods are discussed.

Slides from the 10-minute talk (PDF, 0.4 MB)

Extended version of the talk (PDF, 1.1 MB)