Pushing the Limits of 3D Frequency-Domain FWI with the 2015/2016 OBN Gorgon Dataset

Summary Frequency-domain Full Waveform Inversion (FWI) is suitable to process of full-azimuth/long-offset sparse OBN survey because the inversion can be performed with a few discrete frequencies. Main advantages of the frequency domain are the efficient processing of large number of sources, compact volume of data and cheap implementation of attenuation. However, solving the time-harmonic wave equation in large computational domains (several hundred million grid points) requires cutting-edge linear algebra solver. We apply 3D frequency-domain FWI to the Gorgon OBN data, NorthWestern shelf, Australia. We solve the forward problem with the multifrontal direct solver MUMPS and push FWI up to a frequency of 25 Hz to image a subsurface domain covering an area of 255 km2 down to 8 km depth. The datasets involves 650 nodes processed as reciprocal sources and around 200,000 shots per node. The target is the Gorgon horst between 3.6km and 4.3 km depth, which is separated into blocks by several en-echelon faults. We successfully reconstruct paleo-pockmarks and landslide in the shallow part while the pseudo-reflectivity built by the modulus of the velocity gradient delineates polygonal faults, bounding faults and the intra-horst faults with an unprecedented resolution. Results are validated by seismic modeling and well logs.