Elastic Full Waveform Inversion of Synthetic Data: Marmousi2 Model Test Study
MetadataShow full item record
- Institutt for fysikk 
Full waveform inversion (FWI) is a computationally extensive technique used to obtain estimates of subsurface rock parameters from seismic data. Initial models are iteratively improved by fitting forward modeled data to the acquired, observed data. To reduce the computational cost, a common approach is to execute the forward modeling in the acoustic approximation. Although seismic streamer data consists of received pure P-waves, the acoustic approximation can lead to incorrect modeling through amplitude errors at non-normal incidence reflections and transmissions, and lack of converted waves.This study tests an elastic FWI scheme with normalized least squares misfit minimization on the well-known, geologically complex Marmousi2 model. The inversion was done on the basis of synthetic pressure wave streamer data obtained by finite difference modeling. Inversion has been done for density, S-wave velocity and P-wave velocity. The S-wave velocity and density tests on heavily smoothed initial models gave no improvement, but the FWI scheme recovered the P-wave velocity from a heavily smoothed initial model with great accuracy.Inversion for the P-wave velocity was conducted more comprehensively by experiments based only on transmission data (early arrivals at far offsets) and only on reflection data (near offsets). Equivalent acoustic FWI tests were done to compare the two methods. Independently of the wave phenomena contained in the data, the elastic FWI prove to be superior on streamer data acquired on the Marmousi2 model when the S-wave velocity is assumed to be perfectly known. The main conclusion of this study is that elastic theory is important for the reliability of FWI based on both reflection and on transmission data.The elastic modeling approaches acoustic modeling when the S-wave velocity model is smooth. This means that challenges still remain in how to cope with the S-wave velocity when elastic FWI are used on streamer datasets.