Morphology and depth of reflectors from 2D non-linear inversion of seismic data

Abstract

We present here two methods to obtain reflection images of upper crust seismic reflectors. The techniques are based on migration and waveform coherence analysis of reflected seismic phases recorded in local earthquake seismograms and in active seismic data. The first method is a move-out and stack of reflected seismic phases in local earthquake recordings. The theoretical travel times of reflected/converted phases in a 1D medium for a given interface depth and velocity model are used to align the recordings in time. The locations and origin times of events are initially estimated from the P and S arrival times. Different seismic gathers are obtained for each reflected/converted phase at the interface under consideration, and the best interface depth is chosen as that which maximizes the value of a semblance function computed on moved-out records. This method has been applied to seismic records of microearthquakes that have occurred at the Mt. Vesuvius volcano, and it confirms the reports of an 8- to 10-km-deep seismic discontinuity beneath the volcano that was previously identified as the roof of an extended magmatic sill. The second is a non-linear 2D method for the inversion of reflection travel times aimed at the imaging of a target upper-crust reflector. This method is specifically designed for geophysical investigations in complex geological environments (oil investigations, retrieving of images of volcano structures) where the presence of complex structures makes the standard velocity analysis difficult and degrades the quality of migrated images. Our reflector is represented by nodes of a cubic-spline that are equally spaced at fixed horizontal locations. The method is based on a multiscale approach and uses a global optimization technique (genetic algorithm) that explores the whole of the parameter space, i.e. the interface position nodes. The forward problem (the modelling of reflection travel times) is solved using the finite-difference solver of Podvine & Lecomte (1991) and using an a priori known background velocity model. This non-linear method allows the automated determination of the global minimum (or maximum) without relying on estimates of the gradient of the objective function in the starting model and without making assumptions about the nature of the objective function itself. We have used two types of objective functions. The first is a least-squares L2 norm, defined as the sum of the squared differences between the observed and the calculated travel times. The second is based on coherence measures (semblance). The main advantage of using coherence measures is that they do not require travel-time picking to assess the degree of fit to the data model. Thus, the time performance of the whole procedure is improved and the subjectivity of the human operators in the picking procedure is removed. The methods are tested on synthetic models and have been applied to a subset of data that was collected during the active seismic experiments performed in September 2001 in the gulfs of Naples and Pozzuoli in the framework of what is known as the SERAPIS project.