Data-driven multi-mode interval velocity analysis in terms of P-Q wavefront attributes

We study a data-driven approach to multi-mode interval velocity analysis in laterally inhomogeneous and anisotropic media. We obtain a simple model-independent description of long-spread reflection moveout versus offset for both symmetric and asymmetric wave modes. It resolves a trade-off between anisotropy and lateral heterogeneity. This is done using offset-dependent arrivalangle and wavefront-curvature (P-Q) attributes. These attributes are estimated by transforming common-shot (CS) or common-receiver (CR) data with local slant and beam stacks for various offsets. The relationship between interval parameters and the P-Q attributes is provided by the generalized Dix-type layer-stripping formula. This formula is implemented as a recursive CS or CR wave-equation downward continuation process. The process is performed without reference to an a priori seismic model. We operate with CS or CR gathers without model-based common midpoint (CMP) or common conversion-point (CCP) binning. Our method is robust in that it offers options of wavefield separation and multiple removal in the P-Q Radon-type domain during downward continuation. Examples show that the method is numerically accurate even for large offset-to-depth ratios and in the presence of substantial lateral velocity gradients.