Three-dimensional pre-stack Kirchhoff migration of deep seismic reflection data

This paper describes the implementation of a 3-D pre-stack depth migration and its application to deep seismic reflection data. Based on the Kirchhoff migration approach, the migrated image is generated by weighted integration along diffraction surfaces through the shot record sections. A finite-difference solution of the eikonal equation is used to compute the diffraction surfaces for a given 3-D velocity model. A test on synthetic data and an application to a 3-D deep seismic reflection data set (ISO89-3D), recorded in the vicinity of the German continental deep drill hole (KTB), are detailed. Because of strong lateral velocity contrasts in some parts of this area, the migration was performed in two steps. First, parts of the shot records were migrated using a constant velocity in order to obtain a general overview of the subsurface structure. Afterwards, the whole data set was migrated using a 3-D velocity model derived mainly from wide-angle measurements. Slices through the migrated volume covering an area of about 21 km × 21 km in the horizontal and 15 km in the vertical direction were compared with results from a 2-D survey (KTB8502) carried out within the same region. Over nearly the whole survey area, the SE1 reflector shows up as the continuation of the Franconian Lineament with varying strength of reflection. Its signature, which deviates partly from that of a simple plane reflector, can be followed through the whole upper crust down to about 9 km depth. There, the Erbendorf body appears as a complicated, strongly reflective bowl-shaped pattern with its uppermost top reflection shifted about 5 km to the south-southeast of the KTB drill hole. The results make it clear that the lateral extent and the true depth of these main geological structures (SE1, Erbendorf body) can be interpreted correctly only from the 3-D migration result. Finally, the parallel implementation procedure is described, both for constant- and variable-velocity models.