Initial comparison of 2D, Pseudo-3D and full 3D FWI approaches applied to the Snowflake field data

Anton Ziegon, Jinji Li, Kristopher A. Innanen

Seismic full waveform inversion (FWI) is a powerful tool for the monitoring of CO₂, but full 3D implementations remain computationally expensive. Therefore, we are testing a Pseudo-3D FWI approach that uses coupled 2D FWI frameworks to approximate 3D structure, offering considerable computational savings while improving model consistency across intersections. We compare conventional 2D, full 3D, and this novel Pseudo-3D acoustic FWI approach applied to the Snowflake field data acquired at the Field Research Station (FRS). We detail the acquisition and processing of the Snowflake 1 (baseline) and2 (monitoring) multi-azimuth VSP geophone datasets, including wavefield separation and timelapse matching. A data-driven 3D-to-2D conversion is outlined which is based on the estimation of a complex source wavelet, ultimately improving domain conversion without manual tuning. Initial velocity models are derived from log data and interpolated into the 3D volume as well as 12 intersecting 2D sections used for inversion. Preliminary results show that Pseudo-3D FWI outperforms standalone 2D FWI in structural consistency and intersection alignment, and aligns with full 3D FWI solutions as well as available log data. Initial time-lapse inversions reveal somewhat consistent baseline-monitoring differences within the Basal Belly River Sandstone formation. While these results are encouraging and they also reveal room for improvement, which lead to the future objectives outlined at the end of the report.