Inversion

The Inversion category contains modules that derive subsurface physical properties directly from seismic waveform data. Unlike conventional velocity analysis, which compares moveout patterns on gathers, inversion methods iteratively minimise the difference between observed seismic data and synthetics computed from a parameterised Earth model. The result is a quantitative model — typically a velocity field — whose predicted waveforms match the measured data as closely as possible.

Inversion workflows in g-Platform are designed for depth-domain processing. They require a starting model (commonly produced by tomographic analysis or depth-converted RMS velocities) and observed seismic data that has been pre-processed to suppress acquisition noise and multiples. The modules in this category perform computationally intensive iterative optimisation and support execution on both CPU and CUDA-capable GPU hardware, as well as distributed processing across a g-Platform cluster.

Modules in this category

FWI 2D — Full Waveform Inversion for two-dimensional post-stack seismic data. The module runs a finite-difference exploding-reflector forward model through a depth-domain velocity model, computes a misfit gradient between the synthetic and observed stacks using the Fletcher-Reeves conjugate-gradient method, and iteratively updates the velocity model in slowness space. Three gradient computation strategies are available: Classic (adjoint-state RTM cross-correlation), Spike deconvolution (OMP sparse deconvolution applied to the migrated image), and Integration (deconvolution in time followed by velocity-scaled integration). FWI 2D is the primary tool when high-resolution velocity estimation is required beyond the limits of semblance picking or reflection tomography.