Evolutionary 3D Vs crustal model for Central Apennines

Detailed 3D elastic crustal models are of fundamental relevance to many applications in Geosciences, from geodynamic modelling, to simulation of seismic wave propagation and seismic engineering. However, most of the recent models suffer from two main drawbacks: (1) they are often obtained from interpolation of local 1D models; and (2) they cannot be easily updated, without recomputing the entire model (which, as a corollary, implies the availability of the complete data-set of raw seismic data). The first drawback leads to mainly oversmoothed models on the horizontal scale, where vertical boundaries are not considered either in the 1D models and in the interpolated 3D model. The second disadvantage implies that current models are generally "static" and their updates require a research effort which is often not paying back in terms of outputs.

In this study, we build the framework for an evolutionary elastic model of the Central Apennines. The starting data are represented by a huge data-set of local 1D S-wave velocity models (originally obtained from Receiver Function inversion). We invert such data-set following a Bayesian fusion approach, where the full posterior probability distribution (PPD) of the1D models is exploited to build the 3D elastic model (in absence of the full PPD information, estimators like mean posterior and standard deviation can also be used). The 3D distribution of elastic properties (i.e. a model) is represented by a 3D Voronoi tassellation of the study volume, where the number of 3D Voronoi cells and their positions are unknown. A Markov chain Monte Carlo (McMC) algorithm is used to sample the family of Voronoi models which "fit" the data adequately (here the "data" are the PPD of the 1D models). 

Our results are shown on a regular 5x5x5 km grid down to 100 km depth, and they are consistent with previous models in terms of difference in crustal structure between the Tyrrhenian and Adriatic side of the Apennines. The model shows which of the features are coherent between adjacent stations, and which areas are better resolved. Point of strength over previous models is the possibility of identifying  sub-vertical boundaries, that in a complex region of subduction and neo-formed crust are more likely than a horizontally layered structure. More complementary or additional data (in the form, e.g. of tomographic models or 1D models from dispersion curves) can be easily added to this model, to update it, as new data become available. In fact, new "data" can be either added to the full data-set or can be included modifying the PPD of the 3D Voronoi cells.