Scattering of seismic waves in random velocity models : Finite difference simulation

Abstract: The small-scale heterogeneities within the Earth's crust and the effect these heterogeneities have on seismic waves have gained increasing interest in the last 15 years. One approach to study seismic wave scattering is through numerical simulations in random media. This thesis is mainly concerned with the random media used in scattering simulations. The properties of frequently used random media (Gaussian, exponential andself-similar) are thoroughly investigated with emphasis on the difference between continuous and discrete media. The investigation shows that the standard deviation differs between the two cases and that this should not be corrected for in the modelling. A comparison between the scattering attenuation observed in finite difference simulations and the predictions of single scattering theory implies that it is difficult to estimate the random properties of the scattering medium from seismic data alone and that it maybe difficult to separate scattering and intrinsic attenuation in real data. Two alternative models to the classical random media are presented, aimed to better simulate a fractured crust: (1) a self-similar medium with a lognormal velocity distribution instead of a Gaussian, where the low velocity tail observed in sonic logs from fractured areas is taken intoaccount, and (2) a fracture zone model where the fracture zones are modelled as low velocity zones in a constant high velocity host rock. Properties such as preferred direction and fractal distribution of the widths and lengths of the zones are implemented. Synthetic seismograms produced with finite difference modelling using fracture zone models appear to have more of the characteristics observed in real data than those from the lognormal and Gaussian random models. Scattering simulations of reflection seismicsfrom Iceland, using fracture zone models, indicate that imaging the Moho on low-fold reflection seismic data on Iceland may be difficult due to the scattering in the upper part and random noise levels at later times. Further, in numerical simulations of high resolution reflection seismics, the modelling has to be performed viscoelastically, in some cases with intrinsic Q values as low as 4 in the top 10 m.