A method, comprising: generating a macro P-wave velocity model, including resolution of residual moveout errors, using reflection tomography from a seismic data set; generating a diffraction response from the macro P-wave velocity model for a given common image gather location therein assuming a locally laterally invariant model; converting the generated diffraction response for the given common image gather location to a migrated reflection response to yield a modeled data set; comparing the modeled data set to the given common image point from the seismic data set; updating the macro P-wave velocity model based on the result of the comparison; and iterating the diffraction response generation, the diffraction response conversion, comparison, and update until the modeled data set converges to the common image gathers. This method is an extension of the 1D waveform inversion where the correct source/receiver positions are determined in a locally laterally invariant medium by backprojecting raypaths from each image point using the local dip at that image point.