Using ICA for analysis of seismic events

Independent Component Analysis (ICA) is a statistical and computational technique for revealing hidden factors that underlie set of random variables, measurements, or signals. ICA is a general purpose technique which is used to linearly transform the observed random data into components. The ICA can be estimated by using the concept of maximum nonGaussianity, maximum likelihood estimation, or minimisation of mutual information. This paper applies ICA to seismic acceleration time histories in order to locate any hidden components of ground rotational motion or tilts. Normally the three components of seismically induced rotations are not recorded in most of the available seismic instruments, primarily because previous devices did not provide the required sensitivity to observe rotations in a wide frequency band and distance range (the two horizontal components, equal to tilt at the free surface, are generally recorded at low frequencies) Igel et al 2003. From the x, y and z components usually recorded the Extended Generalised Lambda Distributions (EGLD) – ICA model was used to examine whether rotational or tilt trends were embedded within the 3 components. The algorithm tries to fit a matrix from the data which will separate any other trends within the available components. The results show that the EGLDICA separates trends within the 3 components; however these are not yet identified as tilts or rotations.