HR: 1330h
AN: PP52A-0946    [PDF]
TI: Extrapolation of climatic Time Series based on Principal Component analysis
AU: * Tipenko, G
EM: ffgst@uaf.edu
AF: Geophysical Institute UAF, 903 Koyukuk Dr POB 757320, Fairbanks, AK 99775-7320 United States
AU: Nicolsky, D
EM: ftdjn@uaf.edu
AF: Geophysical Institute UAF, 903 Koyukuk Dr POB 757320, Fairbanks, AK 99775-7320 United States
AU: Romanovsky, V
EM: ffver@uaf.edu
AF: Geophysical Institute UAF, 903 Koyukuk Dr POB 757320, Fairbanks, AK 99775-7320 United States
AB: This research work was directed to study the influence of long-term climate oscillations on permafrost evolution. Paleoclimatic reconstructions can provide boundary conditions for long-term permafrost modeling, whereas there is usually no information about the initial temperature. The use of a steady-state solution reduces uncertainty in initial conditions. To obtain the steady-state temperature distribution it is necessary to extrapolate paleoclimatic reconstructions into the past, which could be done by approximation of available data using trigonometric polynomials.We propose a new high-resolution method of discrete spectrum estimation for time series expressed as a sum of a quasi-periodic trend and noise.The proposed method combines ideas of ESPRIT and of SSA. Principal components (PC) of the time series are determined by an application of an eigenfunction filter to multivariate translated time series. The main observation is that a rotation of elements of PC is stable with respect to noise. A rotation number is proposed as an estimate of the relating frequency of the time series. This number is defined by averaging a rate of the rotation.If noise is taken into account, the extrapolation problem is an ill-posed and non-unique. The current research work is to find the extrapolation of data, using obtained frequencies and the least square approximation to evaluate complex amplitudes. Since the non-harmonics frequencies form an ill-conditioned Gramm matrix, we propose a regularization method that requires the norm of trigonometric polynomial would not increase as a result of extrapolation. The harmonic extrapolation was performed for the historical isotopic temperature record from the Vostok Ice Core. The results of extrapolation were used as boundary conditions for numerical modeling of the long-term permafrost and gas hydrates evolution.
DE: 0674 Signal processing and adaptive antennas
DE: 3230 Numerical solutions
DE: 3260 Inverse theory
SC: Paleoceanography and Paleoclimatology [PP]
MN: 2003 Fall Meeting