Seismic Source Quantitative Parameters Retrieval from InSAR Data and Neural Networks

Abstract

The basic idea of this thesis is to exploit the capabilities of neural networks in a very new framework: the quantitative modelling of the seismic source and the interferogram inversion for retrieving its geometric parameters. The problem can be sum up as follows. When a moderateto- strong earthquake occurs we can apply SAR Interferometry (InSAR) technique to compute a differential interferogram. The latter is used to detect and measure the surface displacement field. The earthquake has been generated by an active, seismogenic, fault having its own specific geometry. Therefore each differential interferogram contains the information concerning the geometry of the seismic source the earthquake comes from; its shape and size, the number of fringes, the lobe orientation and number are the main features of the surface effects field. Two problems have been analysed in this work. The first is the identification of the seismic source mechanism. The second is a typical inversion exercise concerning the fault plane parameter. To perform both exercises of the seismic fault a huge number of synthetic interferograms has been computed. Each of them is generated by a different combination of such geometric parameters. As far as the retrieval of the geometric parameters is concerned an artificial neural network has been properly generated and trained to provide an inversion procedure to single out the geometric parameters of the fault. Five among these latter, Length, Width, Dip, Strike, Depth, have been simultaneously inverted. The result is in agreement with those results based on different approaches. Furthermore the method seems very promising and leads to improve the studies concerning the combined use of neural networks and InSAR technique.