In several branches of seismology and exploration seismics a need exists for fast and accurate raytracing in 3D heterogeneous Earth models to find the (approximate) paths, along which seismic waves travel from source to receiver. Especially earthquake location determination and nonlinear seismic tomography require extensive raytracing for both travel times and ray geometries. Numerous methods (such as bending and shooting methods, finite difference schemes, network or graph searches and perturbation techniques) have been developed for such studies.
In nonlinear seismic tomography, 3D raytracing is an essential part of the iterative solving procedure. The application of very large data sets (on the order of 10 million phases) with many reflected or refracted phases in nonlinear tomography has up to now been hampered by the low computational speed and limited applicability to later arriving phases of existing raytracers.
We present, apply and compare two fast raytracers that are both substantial extensions of existing methods. One is a graph method, based upon the work of Nolet and Moser [1993], the other a perturbation method following the results of Snieder and Sambridge [1993]. We have extended both methods to handle first order (undulating) seismic discontinuities, including (multiple) reflections, head waves and phase conversions. Both methods have been tested with several (synthetic) models to estimate and compare their accuracies and computational speeds. These tests show that the perturbation method traces up to 5 rays per second in a 3D, global Earth model, whereas the graph method takes 5 seconds per ray (on an R8000 processor). The accuracy estimated on the basis of the differences in travel times between the two methods is on the order of 0.15 seconds for global phases with travel times of 1000 seconds or more.
Apart from testing both methods in global models, we will investigate to what extent full 3D raytracing is necessary to obtain accurate paths and travel times in a realistic Earth model. We use the global mantle P wave velocity model obtained by Bijwaard et al. [1998] for this purpose.
Although these raytracers were developed with an application in nonlinear seismic tomography in mind, both methods can just as well be used in other branches of seismology or exploration geophysics.