Observed and simulated inter-decadal changes in the structure of Southern Hemisphere large-scale circulation

Abstract

Several studies have identified that, in the mid-1970s to early 1980s, a major shift occurred in the structure of the large-scale circulation in both hemispheres. This work employs the CSIRO Mk3L general circulation model in ensemble simulations with observed sea surface temperatures (SSTs) and historical time-evolving carbon dioxide (CO₂) concentrations to investigate the inter-decadal changes found observationally in the jet streams, temperature, Hadley circulation, mean sea level pressure and precipitation. First, the performance of the model in simulating these changes for the mean July climate fields of 1949–1968 and 1975–1994, in comparison with the corresponding observations (NCEP/NCAR Reanalysis I and the Twentieth Century Reanalysis V2), is investigated. We find that the model is quite skilful in reproducing the broad features of the important inter-decadal changes that occurred in the mid-1970s. The model simulations and the NCEP/NCAR and twentieth century reanalyses agree in the eastern hemisphere; whereas in the western hemisphere the reanalyses show differences, and the simulations combine aspects of these two datasets. The role of the direct radiative forcing due to CO₂ in driving the inter-decadal changes is also examined. Results indicate that, in comparison with the indirect effect of CO₂ carried by the changing SSTs, there is little additional impact of the direct radiative forcing due to CO₂ on the changes in the latter period. However, our simulations with fixed CO₂ concentration have shown clearly that the atmospheric simulations with historical time-evolving CO₂ concentrations are more skilful in reproducing the inter-decadal changes. The sensitivity of the ensemble results to employing the same or different time evolving sea ice boundary conditions in the ensemble members is also studied. The contributions of internal and external variability are discussed.

Appendix: Clustering Methodology

We use monthly mean HADISST sea surface temperature data over the period January 1870–August 2012. The clustering analysis uses the Finite Element Method (FEM) with Bounded Variation (BV)-regularization and Vector Autoregressive Factor (VARX) models (FEM-BV-VARX) formulated by Horenko (2010) and employed by O’Kane et al. (2013a, b). Full details are given in Metzner et al. (2012) and O’Kane et al. (2013a, b). Briefly the approach assumes that the data set x _t may be represented by a stochastic model with memory effects and by fitting the parameters by a non-stationary stochastic (VARX) model of the form

$$x_{t} = \mu_{\gamma } \left( t \right) + \sum\nolimits_{q = 1}^{m} {A_{\gamma } \left( t \right)x_{t - q\tau } + C_{\gamma } \left( t \right)\varepsilon_{t} }$$

(7)

Here ɛ is a Gaussian process with zero mean, A and C are tensors, μ is the mean, τ is the time lag, and γ denotes the regime state. One might also consider external forcing factors but do not consider them here. The optimal number of cluster states K is determined on the basis of information theory through the Akaike Information Criterion (Horenko 2010; O’Kane et al. 2013a, b). The method simultaneously estimates the clusters (corresponding to regimes) and the most likely meta-stable state transitions between the clusters through the minimization of an average clustering functional L of a given high-dimensional time series. This approach assumes that the dynamics of the observed variable of interest x _t is influenced by the previous m time-lagged values of the same variable (to describe the memory effects) and an unobservable (hidden) stochastic variable ɛ associated with regime transitions that strongly influences the observed variable. Before applying the clustering algorithm we decompose the HADISST anomalies (seasonal cycle subtracted) into Empirical Orthogonal Functions (EOFs) and then use the leading 10 EOFs for the clustering.