Prestack seismic waveform inversion is a highly challenging task. The nonlinearity and nonuniqueness together with computed intensive forward modeling make the problem intractable. Here we report on a new approach to seismic waveform inversion based on a gradient descent scheme that makes use of innovative gradient calculation, adaptive regularization, and an efficient conjugate gradient algorithm. We exploit sparsity of the gradient matrix based on physics in designing an algorithm that is computationally very efficient. The results from our approach are highly encouraging in that we are able to invert large 3‐D dataset with nearly 1,000 model parameters fairly rapidly. Not only that, because of the use of adaptive regularization, we are able to track smooth as well as sharp variations in the impedance, Vp and Vs profiles when realistic starting solutions are used. We have applied the technique successfully to numerous 3‐D exploration and academic datasets. Of particular interest is our ability to estima...
Mrinal K. Sen
Journal of the Acoustical Society of America