Oceanic temperature and salinity are important for studying ocean dynamics and marine ecosystems, and for developing climate prediction models. Physical oceanographers measure these properties using expendable instruments, providing accurate measurements at high vertical resolution at sparse lateral locations. Accuracy of these temperature and salinity data when laterally interpolated and used for ocean dynamics, marine biology, and weather-related studies is questionable. Marine seismic data indicate visible reflections within the water columns, which, in turn, are primarily controlled by the vertical and lateral variations of the ocean water sound speeds. Because these sound-speed variations are functions of temperature and salinity, water-column seismic reflection data have been inverted for sound speeds and used to estimate temperature and salinity. These seismically derived temperature and salinity data have lower vertical, but much finer lateral resolution than those measured by expendable instruments, and therefore are useful for physical oceanography, marine biology, and climate research. However, quality of the seismic inversion relies on a good initial model, which requires generating models at sparse locations over 2D or 3D seismic data volumes and interpolating them over the horizon, manually picked from the stacked seismic data. Because horizon picking is subject to human error and bias, replacing it with an automated process is desirable. We outline a new attribute-guided methodology for initial model generation where no horizon picking is necessary. Starting with this initial model, we run prestack waveform inversion based on genetic algorithm optimization and demonstrate that our method can predict reliable sound-speed profiles. We conclude that our method is a good way for estimating ocean water sound speeds and relating them to the temperature and salinity.
展开▼
