Empirical prognostic models from spatially distributed time series: separation of spatio-temporal modes on the base of M-SSA

It is well known that real systems nearly always are complex and space-distributed and these circumstances complicate the problem of empirical modeling of such systems. One of the ways to reduce the complexity is decomposition of the investigated system into well separated spatio-temporal patterns (modes) that evolve with essentially different time scales. In the report we discuss the new method of decomposition of the Earths climate system into climatic modes. The method is based on generalization of the M-SSA (Multichannel Singular Spectral Analysis) [1] for expanding space-distributed time series in basis of spatio-temporal empirical orthogonal functions (ST-EOF), which taken into account time lag correlations of the processes recorded in spatially separated points. The method is applied to decomposition of the Earths climate system: on the base of 156 years time series of SST anomalies distributed over the globe (http://iridl.ldeo.columbia.edu/SOURCES/.KAPLAN/.EXTENDED/.v2/.ssta/) two climatic modes with essentially different time scales (3-5 and 9-11 years) are separated. For more accurate exclusion of slow (and thus not represented correctly) processes from real data we used ST-EOF basis obtained from time series produced numerically by the INM RAS Coupled Climate Model (http://83.149.207.89/GCM_DATA_PLOTTING/GCM_INM_DATA_XY_en.htm).