A two-step impedance inversion integrating domain adaptation and AVO analysis for the area away from wells

Abstract

Amplitude variation with offset (AVO) inversion is a quantitative seismic interpretation technique used to determine geological formation properties from seismic data and well logs. However, because seismic data do not include low-frequency components, AVO inversion requires the use of additional low-frequency models (LFMs). Generally, LFMs can be generated with high accuracy in the area near wells, but the accuracy of the LFMs decreases away from wells, which influences overall AVO inversion accuracy. To overcome this problem, we propose a two-step approach that combines machine learning with simultaneous inversion to improve AVO inversion accuracy, in regions both near and far from wells. In the first step, a machine learning-based domain adaptation technique is used to predict the acoustic impedance from the well to distant areas. In the second step, the predicted acoustic impedance values are used to construct LFMs that account for the AVO response. These newly generated LFMs are used as the initial model for simultaneous inverse computation. To validate the proposed two-step approach, we applied the approach to field data from the Carnarvon basin in Western Australia. The results demonstrate the feasibility of our approach in obtaining high-accuracy acoustic impedance, shear impedance, and density from the entire seismic dataset, including for regions far from wells.