Diagnosing Changes in ENSO Variability and Predictability from Observations and Models

Abstract:

Recently, it has been suggested that the predictability of ENSO has changed in the last few decades, with reduced forecast skill in the 2000’s relative to earlier decades. We investigate this issue within the context of a statistically stationary stochastically forced multivariate linear dynamical system. Using a Linear Inverse Model (LIM) constructed from observed tropical Pacific oceanic and atmospheric anomalies, it is suggested that decadal changes in realized ENSO forecast skill (for both the LIM and the forecast models of the North American Multi Model Ensemble) may also reflect variations of noise and not necessarily a change in the underlying predictability of tropical Pacific coupled dynamics. That is, some ENSO events are inherently less predictable than others due to the relative importance of different physical processes (eg, surface or thermocline feedbacks) as initially excited by random weather forcing, and periods when such events are more often randomly excited will be periods of reduced forecast skill. A contrast is also drawn with a marked change in (perfect model) tropical Pacific predictability found between two extended simulations of the NCAR CESM1 coupled model, run under fixed 1850 and 2000 radiative forcing conditions, respectively. Diagnosis of surface and thermocline feedbacks in these models (as well as some CMIP5 models) shows how the potential ENSO predictability of some models may appear more sensitive to certain base state changes due to errors in these feedbacks. Finally, the observations-based LIM is used to show that the weak warm event evolution during 2014 was an expected consequence of a tropical Pacific thermocline anomaly that projected weakly on the optimal ENSO precursor, despite the corresponding relatively large equatorial Pacific domain-integrated warm water volume anomaly. In fact, repeating the 2014 LIM forecast by replacing the initial thermocline anomaly with that of spring 1997 yields pronounced warming. In contrast, initial conditions this year are much more consistent with a potentially large, and potentially predictable, 2015/16 event.