Inversão sísmica da forma de onda completa com funcionais baseados no traço analítico

Abstract

Currently, the greatest challenge to overcome in order to apply FWI is avoiding cycle-skipping. Accomplishing this goal requires an initial velocity model that has a significant amount of longwavelength information from the subsurface velocity distribution. In order to produce a suitable initial model, this thesis proposes an inversion approach composed of three stages. The output of each stage is the input of the next one. In the first stage, Dix-Stage, we build a model derived from a coarse Dix velocity analysis. In the second stage, K-Stage, we derive velocity models that honor the first-arrival traveltimes using a transmission inversion approach. In the third stage, FWI-Stage, a concluding waveform inversion process composed by three steps is carried out. The focus here is on the second stage, where we compare first-arrival traveltime tomography with the inversion of a misfit functional based on the windowed early arrivals of the envelope in producing starting models for the FWI-Stage, and on the third stage. We use the Marmousi and Overthrust models to show that, in the K-Stage, the early-arrival envelope functional produces solutions with at least the same quality as those derived from first-arrival traveltime tomography. As a result, in the FWI-Stage, final velocity models with similar quality are obtained not mattering whether this stage was initialized with early-arrival envelope or with first-arrival traveltime tomography models. However, unlike the first-arrival traveltime tomography, the early-arrival envelope inversion does not require time picking, which could save an expressive amount of time. In addition, the early-arrival envelope functional is robust to random noise. We conclude then that the earlyarrival envelope functional has good potential to substitute first-arrival traveltime tomography in the K-Stage. However, neither early-arrival envelope or traveltime functionals are free from problems and, as a result, the interpreter ultimately must judge the resulting velocity models to decide whether they are valid to initialize an FWI process. For the FWI-Stage, we propose a sequence of different misfit functions in the time domain. The employed misfit functions are sensitive to different wavelengths but use the same modeling algorithm and the same model grid. The first step promotes a smooth transition between the K-Stage and the second step, by using the envelope of the complete waveform. In the first inversion step the reflected (late arrivals) component of the wavefield envelopes is fitted, in contrast with the K-Stage that fits the transmitted (early arrivals) component. The phases of the reflected waves are mostly fitted in the third step, which is based on the waveform misfit function preserving only the late arrivals. The second and third steps are of crucial importance to fit the reflected events. We test this approach with the Marmousi model using datasets with different frequencies, obtaining better estimates of the velocity field than those obtained with the classic FWI. The solutions obtained with both classic FWI and sequential inversion approach degrade with a progressively higher peak frequency dataset, but the classic FWI solution degrades more rapidly.