ENSO's non-stationary and non-Gaussian character: the role of climate shifts

Abstract

Abstract. El Niño Southern Oscillation (ENSO) is the dominant mode of climate variability in the Pacific, having socio-economic impacts on surrounding regions. ENSO exhibits significant modulation on decadal to inter-decadal time scales which is related to changes in its characteristics (onset, amplitude, frequency, propagation, and predictability). Some of these characteristics tend to be overlooked in ENSO studies, such as its asymmetry (the number and amplitude of warm and cold events are not equal) and the deviation of its statistics from those of the Gaussian distribution. These properties could be related to the ability of the current generation of coupled models to predict ENSO and its modulation.Here, ENSO's non-Gaussian nature and asymmetry are diagnosed from in situ data and a variety of models (from intermediate complexity models to full-physics coupled general circulation models (CGCMs)) using robust statistical tools initially designed for financial mathematics studies. In particular α-stable laws are used as theoretical background material to measure (and quantify) the non-Gaussian character of ENSO time series and to estimate the skill of "naïve" statistical models in producing deviation from Gaussian laws and asymmetry. The former are based on non-stationary processes dominated by abrupt changes in mean state and empirical variance. It is shown that the α-stable character of ENSO may result from the presence of climate shifts in the time series. Also, cool (warm) periods are associated with ENSO statistics having a stronger (weaker) tendency towards Gaussianity and lower (greater) asymmetry. This supports the hypothesis of ENSO being rectified by changes in mean state through nonlinear processes. The relationship betweenCorrespondence to: J. Boucharel (julien.boucharel@legos.obs-mip.fr) changes in mean state and nonlinearity (skewness) is further investigated both in the Zebiak and Cane (1987)'s model and the models of the Intergovernmental Panel for Climate Change (IPCC). Whereas there is a clear relationship in all models between ENSO asymmetry (as measured by skewness or nonlinear advection) and changes in mean state, they exhibit a variety of behaviour with regard to α-stability. This suggests that the dynamics associated with climate shifts and the occurrence of extreme events involve higher-order statistical moments that cannot be accounted for solely by nonlinear advection.

ENSO regimes: Reinterpreting the canonical and Modoki El Niño

“…The regime of extreme warm events is illustrated in nature by the 1982–83 and 1997–98 events and, to a lesser extent, by the 1877–78 event. Focusing the study of ENSO towards this extreme regime suggests that El Niño is episodic rather than cyclic [ Kessler , 2002] and theoretical work along this line has led to simple models that produce a bursting behavior and asymmetry [ Timmermann et al , 2003; Schopf and Burgman , 2006; Boucharel et al , 2009]. However, the perspective of an episodic El Niño does not consider the moderate regime, and neither do the simple models based on this view.…”

Low‐frequency modulation and trend of the relationship between ENSO and precipitation along the northern to centre Peruvian Pacific coast

“…In addition, ENSO is a non‐stationary phenomenon (Boucharel et al ., ) having a significant modulation of its characteristics (amplitude, frequency and asymmetry) at decadal timescales (Guilderson and Schrag, ; Torrence and Compo, ). For instance, there has been a climate shift in the 1970s from when EP El Niño events became stronger (Miller et al ., ).…”

Can we distinguish canonical El Niño from Modoki?

“…Common diagnostic methods in climate research such as simple box averaging, threshold‐based compositing, linear regression, and principal component analysis ask phenomena to fit a mold that is linear, stationary, and/or symmetric. ENSO is known to defy all of these assumptions [ An and Jin , ; Boucharel et al ., ; Okumura and Deser , ]. Further, such methods are necessary for generalizing large numbers of realizations.…”