Statistical methods

1995
Plaut, Guy, Michael Ghil, and Robert Vautard. “Interannual and Interdecadal Variability in 335 Years of Central England Temperatures.” Science 268, no. 5211 (1995): 710–713. Abstract

Understanding the natural variability of climate is important for predicting its near-term evolution. Models of the oceans' thermohaline and wind-driven circulation show low-frequency oscillations. Long instrumental records can help validate the oscillatory behavior of these models. Singular spectrum analysis applied to the 335-year-long central England temperature (CET) record has identified climate oscillations with interannual (7- to 8-year) and interdecadal (15- and 25-year) periods, probably related to the North Atlantic's wind-driven and thermohaline circulation, respectively. Statistical prediction of oscillatory variability shows CETs decreasing toward the end of this decade and rising again into the middle of the next.

Jiang, N., J. David Neelin, and Michael Ghil. “Quasi-quadrennial and quasi-biennial variability in the equatorial Pacific.” Climate Dynamics 12 (1995): 101–112. Abstract

Evaluation of competing El Niño/Southern Oscillation (ENSO) theories requires one to identify separate spectral peaks in equatorial wind and sea-surface temperature (SST) time series. To sharpen this identification, we examine the seasonal-to-interannual variability of these fields by the data-adaptive method of multi-channel singular spectrum analysis (M-SSA). M-SSA is applied to the equatorial band (4°N-4°S), using 1950 1990 data from the Comprehensive Ocean and Atmosphere Data Set. Two major interannual oscillations are found in the equatorial SST and surface zonal wind fields, U. The main peak is centered at about 52-months; we refer to it as the quasi-quadrennial (QQ) mode. Quasi-biennial (QB) variability is split between two modes, with periods near 28 months and 24 months. A faster, 15-month oscillation has smaller amplitude. The QQ mode dominates the variance and has the most distinct spectral peak. In time-longitude reconstructions of this mode, the SST has the form of a standing oscillation in the eastern equatorial Pacific, while the U-field is dominated by a standing oscillation pattern in the western Pacific and exhibits also slight eastward propagation in the central and western Pacific. The locations of maximum anomalies in both QB modes are similar to those of the QQ mode. Slight westward migration in SST, across the eastern and central, and eastward propagation of U, across the western and central Pacific, are found. The significant wind anomaly covers a smaller region than for the QQ. The QQ and QB modes together represent the ENSO variability well and interfere constructively during major events. The sharper definition of the QQ spectral peak and its dominance are consistent with the “devil's staircase” interaction mechanism between the annual cycle and ENSO.

1993
Penland, Cécile, and Michael Ghil. “Forecasting Northern Hemisphere 700\mbox-mb geopotential height anomalies using empirical normal modes.” Monthly Weather Review 121, no. 8 (1993): 2355–2372. Abstract
Multivariate linear prediction based on single-lag inverse modeling is developed further and critically examined. The method is applied to the National Meteorological Center analyses of Northern Hemisphere 700-mb geopotential height anomalies, which have been filtered to eliminate periods shorter than 10 days. Empirically derived normal modes of the randomly forced linear system are usually correlated, even at zero lag, suggesting that combinations of modes should be used in predictions. Due to nonlinearities in the dynamics and the neglect of interactions with other pressure levels, the lag at which the analysis is performed is crucial; best predictions obtain when the autocovariances involved in the analysis are calculated at a lag comparable to the exponential decay times of the modes. Errors in prediction have a significant seasonal dependence, indicating that the annual cycle affects the higher-order statistics of the field. Optimized linear predictions using this method are useful for about half a day longer than predictions made by persistence. Conditional probabilities are much more efficiently calculated using normal-mode parameters than from histograms, and yield similar results. Maps of the model's Fourier spectra—integrated over specified frequency intervals and consistent with the assumptions made in a linear analysis—agree with maps obtained from fast Fourier transforms of the data.
1992
Keppenne, Christian L., and Michael Ghil. “Adaptive filtering and prediction of the Southern Oscillation index.” Journal of Geophysical Research: Atmospheres 97, no. D18 (1992): 20449–20454.
PDF
Vautard, Robert, Pascal Yiou, and Michael Ghil. “Singular-spectrum analysis: A toolkit for short, noisy chaotic signals.” Physica D 58, no. 1–4 (1992): 95–126. Abstract
Singular-spectrum analysis (SSA) is developed further, based on experience with applications to geophysical time series. It is shown that SSA provides a crude but robust approximation of strange attractors by tori, in the presence of noise. The method works well for short, noisy time series. The lagged-covariance matrix of the processes studied is the basis of SSA. We select subsets of eigenelements and associated principal components (PCs) in order to provide (i) a noise-reduction algorithm, (ii) a detrending algorithm, and (iii) an algorithm for the identification of oscillatory components. Reconstructed components (RCs) are developed to provide optimal reconstruction of a dynamic process at precise epochs, rather than averaged over the window length of the analysis. SSA is combined with advanced spectral-analysis methods - the maximum entropy method (MEM) and the multi-taper method (MTM) - to refine the interpretation of oscillatory behavior. A combined SSA-MEM method is also used for the prediction of selected subsets of RCs. The entire toolkit is validated against a set of four prescribed time series generated by known processes, quasi-periodic or chaotic. It is also applied to a time series of global surface air temperatures, 130 years long, which has attracted considerable attention in the context of the global warming issue and provides a severe test for noise reduction and prediction.
1991
Ghil, Michael, and Robert Vautard. “Interdecadal oscillations and the warming trend in global temperature time series.” Nature 350, no. 6316 (1991): 324–327. Abstract

The ability to distinguish a warming trend from natural variability is critical for an understanding of the climatic response to increasing greenhouse-gas concentrations. Here we use singular spectrum analysis1 to analyse the time series of global surface air tem-peratures for the past 135 years2, allowing a secular warming trend and a small number of oscillatory modes to be separated from the noise. The trend is flat until 1910, with an increase of 0.4 °C since then. The oscillations exhibit interdecadal periods of 21 and 16 years, and interannual periods of 6 and 5 years. The interannual oscillations are probably related to global aspects of the El Niño-Southern Oscillation (ENSO) phenomenon3. The interdecadal oscillations could be associated with changes in the extratropical ocean circulation4. The oscillatory components have combined (peak-to-peak) amplitudes of >0.2 °C, and therefore limit our ability to predict whether the inferred secular warming trend of 0.005 °Cyr-1 will continue. This could postpone incontrovertible detection of the greenhouse warming signal for one or two decades.

1989
Vautard, Robert, and Michael Ghil. “Singular spectrum analysis in nonlinear dynamics, with applications to paleoclimatic time series.” Physica D 35, no. 3 (1989): 395–424. Abstract

We distinguish between two dimensions of a dynamical system given by experimental time series. Statistical dimension gives a theoretical upper bound for the minimal number of degrees of freedom required to describe tje attractor up to the accuracy of the data, taking into account sampling and noise problems. The dynamical dimension is the intrinsic dimension of the attractor and does not depend on the quality of the data. Singular Spectrum Analysis (SSA) provides estimates of the statistical dimension. SSA also describes the main physical phenomena reflected by the data. It gives adaptive spectral filters associated with the dominant oscillations of the system and clarifies the noise characteristics of the data. We apply SSA to four paleoclimatic records. The principal climatic oscillations, and the regime changes in their amplitude are detected. About 10 degrees of freedom are statistically significant in the data. Large noise and insufficient sample length do not allow reliable estimates of the dynamical dimension.

Pages