고해상 탄성파 자료 보정과 향상 및 쇄설성 박층 저류암의 지오모델링

Abstract

This dissertation consists of three self-contained chapters: (1) correction for profile-length water-column height variations in high-resolution seismic data, (2) enhancement of sparker data for stratigraphic interpretation, and (3) geomodeling of a thin clastic reservoir: the Second Wall Creek Sand, Teapot Dome field, Wyoming, USA. The first chapter presents a simple method to correct long-wavelength water-column height variations in shallow-water, high-resolution seismic data. First, the seafloor depth from seismic data is subtracted from the bathymetric grid at the locations where the seismic shot points and the bathymetric grid points are collocated or closest. Then, the result is gridded and smoothed, and extracted for each seismic profile. Mis-tie corrections are added to the extracted values. These water-column height variations are converted into two-way traveltimes and loaded to the trace header of seismic data as a total static. I applied this method to the sparker data acquired from the shallow-water area off the western part of Korea where the tidal range is over 7 m. Large corrections between and near the islands are probably due to the amplification and shortening in tidal wavelength caused by rapid shoaling toward the islands. The second chapter proposes a simple processing flow to enhance the quality of sparker data for better stratigraphic interpretation. The processing flow includes, among others, swell filtering, wavelet processing to convert the data into minimum phase, predictive deconvolution to attenuate both short-path and long-path multiples, and F-X filtering to attenuate random noise. This processing flow was applied to a dip-oriented sparker profile from the shallow-water area off SW Korea. Wavelet processing used a single wavelet for the entire data, extracted from the seafloor reflection, which saved considerable processing time. The processing significantly improved the data quality, allowing us to make detailed sequence stratigraphic interpretation which revealed two Holocene, sixth-order(102-103 years) sequences above the acoustic basement. The lower sequence consists of highstand (HST), falling-stage (FSST), lowstand (LST), and transgressive systems tract (TST). Forced regression in the lower sequence, evidenced by the subaerial unconformity, suggests at least about 30 m of base-level fall. The upper sequence consists only of HST and TST, suggesting no base-level fall. The third chapter is a geomodeling case study optimized for the Second Wall Creek Sand, Teapot Dome field, Wyoming, USA. The Second Wall Creek Sand is a thin (about 20 m), coarsening-upward, fluvial-deltaic reservoir deposited during sealevel lowstand. The resolution of the seismic data was improved by bandwidth enhancement to better pick the top and base of the Second Wall Creek Sand. I treated the Second Wall Creek Sand in the study area as a non-compartmentalized reservoir since fault displacements are very small, giving rise to sand-to-sand juxtaposition. The geomodeling procedure consisted of structural modeling of the depth 3-D grid (30 m by 30 m by 1 m cells) and petrophysical modeling for porosity and net-to-gross (N/G). The Second Wall Creek Sand was assumed to be of a single facies because lithofacies and edge-detection seismic attributes of its top surface do not reveal distinct depositional features. The porosity and N/G models were constructed by multi-attribute neural networks, based on respectively from five and four wells. The absolute and relative impedances from inversion and the 20-Hz, 50-Hz, and 80-Hz isofrequencies from Generalized Spectral Decomposition were used in neural-network training in addition to the amplitude-based attributes. The modeled porosity varies from less than 8% to over 18% with an average of 16.7%. The modeled N/G values range from about 0.2 to over 0.8 with an average of 0.7. Very low porosity and N/G occur along the extreme edges of the study area where the seismic data quality is poor. The porosity and N/G models may not be optimal because the neural-network training was based on a small number of wells and the limited number of the time samples for the thin reservoir interval.