Large scale high-frequency seismic wavefield reconstruction, acquisition via rank minimization and sparsity-promoting source estimation

TitleLarge scale high-frequency seismic wavefield reconstruction, acquisition via rank minimization and sparsity-promoting source estimation
Publication TypeThesis
Year of Publication2020
AuthorsShashin Sharan
UniversityGeorgia Institute of Technology
Thesis Typephd
Keywordscompressed sensing, low-rank, PhD, Sparsity-promoting, wavefield reconstruction

Seismic data reconstruction on a dense periodic grid from seismic data acquired on a coarse grid is a common approach followed by most of the oil & gas companies. This approach allows them to save on operationally challenging and expensive dense seismic data acquisition. Dense seismic data is one of the key requirements for generating high-resolution images of earth's subsurface for exploration and production decisions. Based on the Compressive Sensing (CS) paradigm, low-rank matrix factorization based seismic data reconstruction methods are computationally cheaper and scalable to large datasets in comparison to sparsity-promotion based methods. The sparsity-promotion based methods are based on transformation in certain transform domains that can be computationally expensive for large datasets. Although, low-rank matrix factorization based methods perform well at lower frequencies, their performance degrades at higher frequencies due to increase in rank of approximating matrix. One of the contributions of this thesis is a recursively weighted matrix factorization approach to improve the quality of reconstructed data at higher frequencies. This recursively weighted approach exploits the similarity between adjacent frequency slices. Although, recursively weighted method improves the data reconstruction quality at higher frequencies, it can be computationally expensive for large scale seismic datasets. This is because of the interdependence of frequencies preventing simultaneous reconstruction of frequencies. Another contribution of this thesis is a computationally efficient recursively weighted framework for large scale dataset by parallelizing data reconstruction over rows of low-rank factors of each frequency slices. To reduce the cost and turnaround time of seismic data acquisition simultaneous source acquisition is adapted by the oil and gas industry in last few years. Another contribution of this thesis is a low-rank based method for simultaneous separation and reconstruction of seismic data on a dense periodic grid from large scale seismic data acquired with simultaneous source acquisition. Next part of this thesis focuses on accurate detection of fractures created by hydraulic fracturing in unconventional reservoirs for economical production of oil and gas. Fracturing of rocks during hydraulic fracturing gives rise to microseismic events, which are localized along these fractures. In this work, a sparsity-promoting microseismic source estimation framework is proposed to detect closely spaced microseismic sources along with estimation of their associated source-time functions from noisy microseismic data recorded by receivers along the earth's surface or along monitor wells. Detecting closely spaced microseismic events helps in delineating fractures and estimation of source-time function is useful in estimating fracture's origin in time. Also, source-time functions can be potentially useful for estimating the source-mechanism. Also, this method does not make any prior assumption on number of microseismic sources or shape of their source-time functions. Therefore, this method is useful for detecting microseismic sources with different source signatures and frequency content. Last part of this thesis focuses on sparsity-promoting photoacoustic imaging to detect photoabsorbers along with estimating the associated source-time functions. Traditional photoacoustic imaging can only estimate the locations of photoacoustic absorbers. Also, traditional methods require dense transducer coverage whereas sparsity-promotion based method can work with reduced transducer sampling reducing the overall data storage cost.



Citation Keysharan2020THlsh