Waveform tomography based on local image correlations

Common velocity analysis is based on the invariance of migrated images with respect to the experiment index (shot number, plane-wave take-off angle, etc.) or extension parameters (reflection angle, correlation lags in extended images, etc.). All the information available, i.e. the entire survey, is used for assessing the quality of the model used for imaging. This approach is effective but forces a clear separation between imaging and velocity model building. Here, we ask a complementary question: how much information about the velocity model is contained in a minimum number of images? Starting from an alternative statement of the semblance principle, we propose a measure of velocity error based on local correlations of pairs of migrated images. We design an objective function and implement a local, gradientbased optimization scheme to reconstruct the velocity model. Our methodology is “full-wave” because it is not based on a linearization of the imaging operator (in contrast with linearized wave-equation migration velocity analysis techniques). The gradient of the objective function is computed using the adjoint-state method.