Statistical selection of the 'best' seismic source mechanisms from inversions of synthetic volcanic long-period events

Inversions for the source mechanism of long‐period (LP) seismicity recorded on
volcanoes have become increasingly common and are used to interpret fluid‐driven
processes. The source mechanism considered for LP inversions usually consists of a
symmetric moment tensor with and without single forces. Also, constrained inversions have
been performed where one presumes a specific source geometry that reduces the degrees of
freedom in the inversion. To select the correct solution from the different possible
mechanisms, the Akaike information criterion (AIC) has been used. However, since
AIC performs well only if the inverted model is close to the true model, we tested its
ability to select the correct model in LP inversions. Using synthetic data sets generated on
Mt. Etna, Italy, with a tomography velocity model and the Green’s functions calculated for
a simplified, homogeneous velocity model, we have investigated (1) if any of the inversion
source models can recover the true mechanism and (2) the ability of the Akaike and
Bayesian information criteria (BIC) to select the correct model. Results show that in some
cases it is possible to recover the mechanism but never the source magnitude and that the
BIC is a better measure than the AIC in selecting the true source model, although in
numerous cases both criteria fail to select the correct solution. Therefore, the BIC should
be used as opposed to the AIC if it is necessary to select an appropriate source. Caution
should be used when using the statistical measure in any seismic inversion application.