Data Assimilation Methods in Seismic Inversion

Abstract

Full-waverform inversion (FWI) is a powerful tool for getting high resolution images from seismic data to understand formations and the reservoir characteristics. But the inversion problems usually do not have unique solutions which means several inversion results could fit the same true model. In this case, it is quite important to estimate the uncertainty of the inversion solutions and their accuracy. Moreover, the cycle skipping artifact is the main challenge of getting a robust FWI result. In this thesis, the goal is to study on the FWI and introduce the multiscale method and the ensemble method which are based on the data assimilation method into FWI algorithm. The multiscale method will use different scale of offsets and travel times in iteration steps to prevent the cycle skipping artifacts, and the ensemble method will run multiple models to estimate the uncertainty of the FWI results. The application of this method is first implemented in two different scales of Marmousi model with 2D synthetic seismic data, and then applied on the 2D line NSR-31150R2, Sequence 124 seismic line from the North Sea. It is shown that applying multiscale and ensemble methods in FWI have a satisfied result of estimating the uncertainty for the inversion solutions both in synthetic data and real field data. The cycle skipping artifacts are limited which makes the algorithm more stable.