The Reverse Time Migration Using Stereo-modeling Operators And Its Applications

With the development of seismic exploration, the targets are becomingincreasingly complex, which leads to a higher demand for the efficiency and accuracyof seismic exploration. In this dissertation, a new kind of reverse time migration (RTM)methods, which use not only displacement but also its gradients, are presented. Thiskind of RTM are based on stereo-modeling and proved to have high-accuracy,high-efficiency, low-dispersion and low-numerical anisotropy.Firstly, a12thorder stereo-modeling (12-STEM) method is developed and analysesincluding the theoretical error, stability condition and numerical dispersion relation aregiven in detail. Compared with high order conventional finite difference methods,Lax-Wendroff correction (LWC) methods, and4thorder stereo-modeling method,12-STEM is proved to be the most accurate, low-dispersive and efficient. The resultsshow that the maximum error of the phase velocity is about3%for12-STEM，whereasit’s about15%for12thorder LWC (12-LWC). Under the condition of no visiblenumerical dispersion in numerical modeling of wave equations, the computational speedof12-STEM is roughly8times faster than that of12-LWC. However,12-STEMrequires only about38%of the storage space for the12-LWC for comparable reliability.Secondly, we apply stereo-modeling methods to RTM for the first time. A fewbenchmark tests are given, as well as many applications for2D and3D cases based onsynthetic and field datasets. The numerical results for Marmousi model, Sigsbee2Bdataset, BP dataset,3D SEAM model,3D SEG/EAGE classic A1and NA-C3datasetsshow that, based on the stereo-modeling, the RTM can get high quality images usingcoarse gird sizes and under high frequencies in very complex areas. The field datasetshows improvements to the structure, continuity and positioning in the image obtainedby stereo-modeling compared with the image from LWC.Finally, we test the stereo-modeling using Marmousi model,3D SEG/EAGE saltmodel and Overthrust model to verify its ability to handle the aliasing problems whentrace spacings of datasets are large in cross-line direction. The numerical results showthat stereo-modeling can double the trace spacing or even use larger ones to get thesame RTM results as LWC when gradients are inputted on the surface for stereo-modeling. The quality of the results decreases a little but still satisfying when weinput zero instead of exact gradient information.