Abstract:
The identi cation of the small scale geological heterogeneities like pinchouts,faults ,fractures
etc is a primary problem in seismics.They hold a signi cant importance in understanding
the geology of a particular area as they can act as structural traps for accumulation
of hydrocarbons or can behave as migration pathways aiding the escape of the
hydrocarbons.
Standard approaches to obtain this high-resolution information, such as coherency analysis
and structure-oriented lters, derive attributes from stacked, migrated images are
image-driven.These techniques are sensitive to artifacts due to an inadequate migration
velocity; in fact the attribute derivation is even not based on the physics of wave propagation.
These small scale features are usually encoded in the form of di ractions in our seismic
data.Thus a seismic section containing only di ractions could be of great value to the
interpreter .Di ractions are the other coherent events present along with the re
ections
in our seismic data.
There are some fundamental di erences between re
ections and di ractions like di erent
moveout, amplitude etc which can facilitate there separation and here we discuss two
methods to focus and separate di ractions. The rst part of the thesis is devoted to
exploiting the di erence between di ractions and re
ections on the basis of Snell's law
. We try to utilise this to design a weighting function which account this di erence in
behaviour to directly clean the gradient in Full waveform Inversion.
4
The second part of the thesis proposes separating the di ractions using a dip-steering
lter in the model domain in the Born framework which could be directly inverted using
Full waveform inversion.
The algorithms developed in this thesis are coded using GPU's ( CUDA C ) in a parallel
environment except the dip-steering lter algorithm.
