Reciprocity-based imaging using multiply scattered waves

Abstract

In exploration seismology, seismic waves are emitted into the structurally complex Earth. Its response, consisting of a mixture of arrivals including primary reflections, conversions, multiples, and transmissions, is used to infer the internal structure and properties. Waves that interact multiple times with the inhomogeneities in the medium probe areas of the subsurface that are sometimes inaccessible to singly scattered waves. However, these contributions are notoriously difficult to use for imaging because multiple scattering turns out to be a highly nonlinear process. Conventionally, imaging algorithms assume singly scattered energy dominates data. Hence these require that energy that scatters more than once is attenuated. The principal focus of this thesis is to incorporate the effect of complex nonlinear scattering in the construction of subsurface elastic images. Reciprocity theory is used to establish an exact relation between the full recorded data and the local (zero-offset, zero-time) scattering response in the subsurface which constitutes our image. Fully nonlinear, elastic imaging conditions are shown to lead to better illumination, higher resolution and improved amplitudes in pure-mode imaging. Strikingly it is also observed that when multiple scattering is correctly handled, no converted-wave energy is mapped to any image point. I explain this result by noting that conversions require finite time and space to manifest. The construction of wavefield propagators (Green’s functions) that are used to extrapolate recorded data from the surface to points in the Earth’s interior is a crucial component of any imaging technique. Classical approaches are based on strong assumptions about the propagation direction of recorded data, and their polarization; preliminary steps of wavefield decomposition (directional and modal) are required to extract upward propagating waves at the recording surface and separate different wave modes. These algorithms also generally fail to explain the trajectories of multiply scattered and converted waves, representing a major problem when constructing nonlinear images as we do not know where such energy interacted with the scatterers to be imaged. A secondary aim of this thesis is to improve on the practice of wavefield extrapolation or redatuming by taking advantage of the different nature of multi-component data compared with single-mode acoustic data. Two-way representation theorems are used to define novel formulations in elastic media which allow both up- and downward propagating fields to be back-propagated correctly without ambiguity in the direction, and such that no cross-talk between wave modes is generated. As an application of directional extrapolation, the acoustic counterpart of the new approach is tested on an ocean-bottom cable field dataset acquired over the Volve field, North Sea. Interestingly, the process of redatuming sources to locations beneath a complex overburden by means of multi-dimensional deconvolution also requires preliminary wavefield separation to be successful: I propose to use the two-way convolution-type representation as a way to combine full pressure and particle velocity recordings. Accurate redatumed wavefields can then be obtained directly from multi-component data without separation. Another major challenge in seismic imaging is to construct detailed velocity models through which recorded data will be extrapolated. Nowadays the information contained in the extension of subsurface images along either the time or space axis is commonly exploited by velocity model building techniques acting in the image domain. Recent research has shown that when both extensions are taken into account, it is possible to estimate the data that would have been recorded if a small, local seismic survey was conducted around any image point in the subsurface. I elaborate on the use of nonlinear elastic imaging conditions to construct such so-called extended image gathers: missing events, incorrect amplitudes, and spurious energy generated from the use of only primary arrivals are shown to be mitigated when multiple scattering is included in the migration process. Finally, having access to virtual recordings in the subsurface is also very useful for target-oriented imaging applications. In the context of one-way representation, I apply the novel methodology of Marchenko redatuming to the Volve field dataset as a way to unravel propagation effects in the overburden structure. Constructed wavefields are then used to synthesize local, subsurface reflection responses that are only sensitive to local heterogeneities, and detailed images of target areas of the subsurface are ultimately produced. Overall the findings of this thesis demonstrate that, while incorporating multiply scattered waves as well as multi-component data in imaging may be not a trivial task, such information is vital for achieving high-resolution and true-amplitude seismic imaging.