A method for building a three dimensional (3D) model of a subsurface formation includes selecting, from a set of seismic shots, a plurality of first arrival signals representing the seismic shots. The method includes applying a quality control function to the plurality of first arrival signals to obtain a set of remaining first arrival signals. For each remaining first arrival signals, the method includes applying a velocity inversion function to obtain a depth velocity value at a common-midpoint (CMP) location in a shot gather including the seismic shot associated with that remaining first arrival signal, the CMP location representing a lateral variation of the shot gather including that seismic shot. The method includes, based on the depth velocity value for the seismic shot associated with each remaining first arrival signal, generating a velocity model representing the 3D model of the subsurface formation.