Towards Solving the Mixing Problem in the Decomposition of Geophysical Time Series by Independent Component Analysis

The useof the Principal ComponentAnalysis techniquefor the analysis of geophysicaltime serieshas beenquestionedin particular for its tendency to extract componentsthat mix severalphysical phenomenaevenwhen the signal is just their linear sum. We demonstrate with a data simulation experiment that the Independent ComponentAnalysis, a recently developedtechnique, is able to solve this problem. This new technique requires the statistical independenceof components,a stronger constraint, that useshigher-orderstatistics, instead of the classical decorrelation, a weakerconstraint, that usesonly second-orderstatistics. Furtherm,,re, ICA does not require additional a _priori information such as the localization cot straint used in Rotational Techniques.