Methods for Attenuation of Seismic Ground Roll and 4D Traveltime Inversion Using Diving Waves

Abstract

The thesis comprises methods for seismic data processing, namely seismic ground roll attenuation, as well as a method for 4D traveltime inversion. Ground roll, also known as Rayleigh waves, in land surveys as well as mud roll, also known as Scholte waves, in marine surveys is among the strongest and most troublesome noises. For many reasons, this type of noise typically covers large parts of seismic records, dominating them and, therefore, masking underlying reflection events. Many seismic imaging, processing, and interpretation techniques are modeled as noisefree processes. If ground roll is not accurately subtracted from the original field record, these techniques often fail and, therefore, the interpreter can be led astray by a distorted representation of the subsurface. Various methods exist to suppress ground roll, in both acquisition and processing stages. In this thesis, I have made an overview of processing methods. They fall into two categories: single-station and multistation filters. I have contributed to the development and perfection of both categories. The thesis includes three papers on ground roll attenuation, one of which is devoted to single-station filters, and the other two to multistation filters. Single-station procedures for mitigating ground roll on multicomponent seismic data that are called polarization filters exploit not only elliptic polarization of this type of noise, but also its high amplitudes and low frequencies. I have focused on an SVDbased polarization filter that I consider one of the best in the industry. I have improved its ground-roll detector to make it theoretically insensitive to ambient noise and more sensitive to the presence of ground roll. The advantages of the new detector have been demonstrated on synthetic and field data sets. Also, the damage that can be done by polarization filtering to the underlying reflections have been theoretically estimated and the result has been checked against stochastic and 2.5D 3C finite-difference synthetic data sets. Multistation filters can be applied to single-component and multicomponent data sets. The main advantages of them lie in their ability to take into account the spatial coherence of both ground roll and reflected waves, and the difference in their apparent velocities. A new least-squares method for signal estimation with a complicated and more realistic mathematical model of the multichannel seismic record containing random noise and an arbitrary number of coherent noise wavetrains has been developed. It has been shown that under certain conditions, the method may be reduced to two successive stages, namely preliminary subtraction of estimates of all the coherent noise wavetrains and final estimation of the signal from the residual record. In both stages, optimum weighted stacking is used with reference to the variances of random noise and to the amplitudes and arrival times of the corresponding coherent component. A simplified scheme and an advanced scheme for subtracting coherent noise have been proposed, which are called the zeroth-order and first-order approximations, respectively. Later on, this method has been improved and applied for signal refinement on field data sets contaminated by severe, mildly dispersive ground roll and additive random noise. The merits of the new method over conventional f-k filtering and SVD-based filtering in the presence of appreciable additive random noise whose energy varies significantly from trace to trace have been shown. The thesis also includes a paper devoted to 4D traveltime inversion for quantitative estimation of velocity anomalies. Conventional methods lie in inverting 4D time shifts of waves reflected beneath the target zone into velocity anomaly parameters, such as its thickness and magnitude. In this thesis, an alternative method that implies inverting time shifts of diving waves instead of reflected waves has been suggested. Our motivation is based on the fact that the diving waves have a longer travel path when passing through the horizontal layer compared with the one for reflected from deeper reflectors waves. The background model, which is assumed to be known, is a stack of isotropic layers which is approximated by a linear-with-depth velocity model to initiate diving-wave propagation. In the monitor case a shallow negative velocity anomaly of a box shape appears. It has been demonstrated that 4D time shifts caused by this anomaly which are observed on diving waves are much higher than those observed on the waves reflected from a deep reflector. It has also been shown that the plot of diving-wave time shifts provides more information on the anomaly than that of reflected waves. It has been suggested to extract attributes from the plot of diving-wave time shifts which are then inverted into the anomaly parameters. With a finite-difference data set, it has been demonstrated that these attributes provide us with estimates of the anomaly parameters which are very close to their true values. Thus estimated parameters can be either admitted as final values or used as initial information for time lapse full waveform inversion (4DFWI) or any other inversion algorithm. I have classified the work in this thesis within five main themes: 1. Overview of previous research on ground-roll mitigation. 2. Investigation and perfection of single-station polarization filtering of ground roll. 3. Investigation and perfection of multistation filtering of ground roll. 4. A brief overview of previous research on 4D inversion. 5. Representation and testing of the method for 4D traveltime inversion using diving waves.