IMAGE RECONSTRUCTION OF 2D VELOCITY MODELS IN SEISMIC IMAGING

Abstract

In this thesis, images of 2 D velocity models are reconstructed using Full Waveform Inversion which is based on the acoustic wave equation in both time and frequency domain. I have written MATLAB codes for Full Waveform Inversion (FWI) and then tested on few synthetic models. The main objective was to evaluate the capabilities of FWI to estimate velocities from the acquired synthetic data. Standard seismic imaging tool such as Travel Time Tomography usually fail when applied to a complex geological structure. Indeed, the velocity contrast between the area of interest and the surroundings usually complicates the propagation of the seismic wave, thus making the imaging of the target body quite difficult. Full waveform Inversion (FWI) has shown the capabilities to image even in complicated subsurface in past years and emerged as a valuable seismic imaging tool which is now widely used in exploration industries. In this thesis, firstly, the FWI code is tested on velocity models such as high velocity layer and low velocity layer models. These models represent a very familiar velocity structure found on near surface hence, dealing with these models is pretty reasonable. After successful testing, the FWI code was also tested on complicated Marmousi model in both time and frequency domain. One of the biggest contributions of the Marmousi model is that it demonstrated the limitations of first-arrival travel times in imaging complex media. Specifically, Geoltrain and Brac (1993) showed that multi-arrival travel times are needed in order to properly image the Marmousi model. In portions of the Marmousi model, or any other complex model, the first arrival is not necessarily the most energetic. Therefore, reflected energy from key horizons, such as the top of the reservoir, are not properly imaged by using only first arrival travel times (Alkhalifah, 1998). V The final results were then compared with the real velocity model to analyze the amount of recovered velocities. However, in order to initiate FWI an initial model is required which was derived from smoothing the true velocity model by using Gaussian smoother. This is an easy way to ensure an adequate starting model, as the FWI method is known to be sensitive on starting model. The results in this master thesis demonstrate that image reconstruction done by FWI both in time and frequency domain matches very well with the true model used in this thesis. For the future, applying FWI to real data from more complex geological medium and developing a migration tool and test the effect of FWI on a migrated image, are interesting challenges.