Joint inversion of simultaneous source time-lapse seismic data sets

Cascaded covariance-based preconditioning

We have specialized the spatial and temporal preconditioners such that the dip-discrimination (or range) of the filters decreases as a function of iteration, while the temporal integration leak rate increases as a function of iteration. This preconditioning approach (which should be applicable to other inversion problems) ensures that close to the solution, the data fitting goal is given more weight relative to the regularization goal.

In addition, because non-stationary deconvolution by polynomial division can become unstable at sharp boundaries, the filter range at any image point is a function of dip contrast-dependent covariance. Details of this preconditioning approach is outside the scope of this paper and will be discussed elsewhere.