A stochastic perspective on atmospheric regime behavior is outlined, based on a treatment of climate variability as a stochastic system with state dependent noise.

The existence of persistent atmospheric flow regimes is of central importance in assessing long-range predictability. Multiple climate regimes are believed to be due to nonlinearities in the equations governing atmospheric dynamics. Climate regimes are typically studied by examining the bivariate probability distribution describing the two leading Empirical orthogonal functions of appropriate atmospheric variables. Ideally, one would like to find significant multiple peaks in these PDFs. However, observed PDFs very seldom show any clear multimodality at all. Rather, PDF "inhomogeneities" (deviations from bivariate Gaussianity) are often interpreted as multiple regimes resulting from nonlinear atmospheric dynamics.

There is one outstanding issue with this interpretation. If there are indeed multiple regimes in the internal atmospheric dynamical system, are they the result of predictable nonlinear behavior, or are they the result of unpredictable forcing by state-dependent noise? We show in an analysis of 750 hPa streamfunction data, that the non-Gaussian regime behavior in the leading EOFs is not induced by nonlinearities in the deterministic part of the motion, but is rather due to multiplicative noise. Thus, non-Gaussianity does not always imply that a system has nonlinear multiple regimes.