Local high-resolution passive seismic tomography and Kohonen neural networks — Application at the Rio-Antirio Strait, central Greece

Abstract

A high-resolution passive seismic investigation was performed in a [Formula: see text] area around the Rio-Antirio Strait in central Greece using natural microearthquakes recorded during three months by a dense, temporary seismic network consisting of 70 three-component surface stations. This work was part of the investigation for a planned underwater rail tunnel, and it gives us the opportunity to investigate the potential of this methodology. First, 150 well-located earthquake events were selected to compute a minimum (1D) velocity model for the region. Next, the 1D model served as the initial model for nonlinear inversion for a 3D P- and S- velocity crustal structure by iteratively solving the coupled hypocenter-velocity problem using a least-squares method. The retrieved [Formula: see text] and [Formula: see text] images were used as an input to Kohonen self-organizing maps (SOMs) to identify, systematically and objectively, the prominent lithologies in the region. SOMs are unsupervised artificial neural networks that map the input space into clusters in a topological form whose organization is related to trends in the input data. This analysis revealed the existence of five major clusters, one of which may be related to the existence of an evaporite body not shown in the conventional seismic tomography velocity volumes. The survey results provide, for the first time, a 3D model of the subsurface in and around the Rio-Antirio Strait. It is the first time that passive seismic tomography is used together with SOM methodologies at this scale, thus revealing the method’s potential.