Spectral analysis of uneven time series of geological variables

Abstract

In geosciences the sampling of a time series tends to afford uneven results, sometimes because the sampling itself is random or because of hiatuses or even completely missing data or due to difficulties involved in the conversion of data from a spatial to a time scale when the sedimentation rate was not constant. Whatever the case, the best solution does not lie in interpolation but rather in resorting to a method that deals with the irregular data. We show here how the use of the smoothed Lomb-Scargle periodogram is both a practical and efficient choice. We describe the effects on the estimated power spectrum of the type of irregular sampling, the number of data, interpolation, and the presence of drift. We propose the permutation test as being an efficient way of calculating statistical confidence levels. By applying the Lomb-Scargle periodogram to a synthetic series with a known spectral content we are able to confirm the validity of this method in the face of the difficulties mentioned above. A case study with real data, including hiatuses, representing the thickness of the annual banding in a stalagmite, is chosen to demonstrate an application using the statistical and physical interpretation of spectral peaks.