Application of full waveform inversion to short-offset seismic data lacking low frequencies
Abstract
Adjoint-state full waveform inversion (FWI) is an HPC technique that potentially allows extracting models of the rocks elastic properties with a resolution that is higher than that of almost any other geophysical method. Despite its potential, FWI suffers from a number of issues that makes it challenging to be applied to most types of field data. One of the main problems is the high non-linearity, which makes the solution strongly dependent on the initial model selected. In particular, for the commonly-used L2-norm objective function, successful FWI requires initial models that allow reproducing seismograms differing by less than half a period from the recorded ones. A typical solution is applying arrival travel-time tomography (TTT) in a first stage and use the resulting model as initial one for FWI. However, this approach does not work when the original data lack low frequencies and have only short offsets, as it occurs with streamer data in deep water settings, so that there are no first arrival refractions in the original recordings. In this work, we present a number of solutions that allow overcoming these issues and apply FWI to short-offset streamer data lacking low frequencies. These include (1) the wave equation downward continuation (DC) of the streamer data followed by joint refraction and reflection TTT to obtain a suitable starting model for FWI, and (2) using robust, instantaneous phase-based functions instead of the typical L2-norm to compare simulated and recorded data. We show that both approaches allow successful application of FWI to limited-offset data (<10 km) starting at frequencies of >4 Hz, typically present in field data recordings. In particular, we present field data examples showing that the combination of DC, joint refraction and reflection TTT, and FWI, allows obtaining accurate high-resolution velocity models using data acquired with a 480 channel, 6 km-long streamer in a >2 km deep water layer, starting FWI at a frequency as high as 8 Hz. The obtained FWI velocity model allows identifying a number of geologically meaningful details, as a high-velocity salt layer interbedded within sediments, which can not be identified otherwise
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMDI24B..26S
- Keywords:
-
- 0545 Modeling;
- COMPUTATIONAL GEOPHYSICSDE: 0560 Numerical solutions;
- COMPUTATIONAL GEOPHYSICSDE: 1932 High-performance computing;
- INFORMATICSDE: 3260 Inverse theory;
- MATHEMATICAL GEOPHYSICS