Interpretation of seismic data in terms of anisotropy

Abstract

A lot of seismic data present features which remain unexplained or poorly explained after an interpretation is made in terms of classical isotropic models. Seismologists are used to employing the word “noise” to explain why their models do not fit the data better. Even if part of the misfit is recognised as of deterministic nature, it is often small enough to be considered as due to some second order effects which are not taken into account in the theory. The problem is that these “second order effects” are sometimes not so small… It is then worthwhile to have a closer look at the data, be they parametric, such as arrival times or dispersion curves, or non-parametric, such as waveforms or amplitude spectra. Let us take the example of teleseismic P-wave arrival times measured on dense networks of seismological stations. Papers published on tomographic inversions of such parametric data quite commonly explain less than 50% of the initial variance of the observations. Very small variance reductions were indeed quite common in the earliest large-scale tomographic studies based on bulletin data. In Europe for example, the three-dimensional P-velocity model of Hovland and Husebye (1981) explained only 23% of the initial data variance while Romanowicz (1980) explained 33% of the variance for a similar data set. Errors in the readings of the arrival-times are often cited as the main source of noise in such data. In fact, even if the readings are not taken from the station bulletins but are controlled on the records themselves, the unexplained variance in the data can remain quite large.