Spectral analysis with incomplete time series: an example from seismology

A method for spectral analysis of nonequidistantly spaced time series is presented: the CLEAN algorithm performs an iterative deconvolution of the spectral window in the frequency domain. We demonstrate the capability of the method on synthetic data examples and apply CLEAN to seismological data, in an example where we seek temporal changes in elastic wave velocities. The observed periodic changes of phase differences consist of frequencies, which in principle can be explained by the influence of solid earth tides, but also by other effects with similar periodicities. Only CLEAN enabled us to enlarge the time window over missing data segments until the frequency resolution was accurate enough to rule out solid earth tides as cause for the observed periodic changes. A MATLAB version of the CLEAN algorithm is available from the authors, or from the IAMG server, (C) 1999 Elsevier Science Ltd. All rights reserved.