Filtering of Multivariate Time Series in the Principal Component Space

Ludwik Liszka
Swedish Institute of Space Physics
Sörfors 634
S-905 88 Umeå


Observing a non-stationary physical phenomenon, a time series of a variable is measured. Usually, there are several variables which must be measured in order to get complete information about the process. The experiment results in a set of univariate time series. The question is whether the conventional method of handling time series (each one separately), reveals all the properties of the process.

An alternative method to approach the problem is to treat all the observed time series as a single multivariate time series. The principal component method is one of possibilities to transform the multivariate time series into a different space, usually of less dimensions. The principal component method offers an interesting possibility to search for periodical or quasi-periodical phenomena in the data through filtering of multivariate data in the principal component space.

The present report shows examples of filtering of multivariate data in the principal component space: enhancement of waves in multisensor temperature measurements and removal of instrumental effects in a sequence of frequency spectra.

The filtering in the principal component space together with decomposition techniques improves our possibilities to study complex processes.

IRF Scientific Report 238

January 1997.

Download report

Back to previous page

HTML by Peje Nilsson ( peje_&_irf_dot_se )
This page was updated Wednesday, 23-Dec-2020 14:29:02 CET.

Valid XHTML 1.0!

Valid CSS!