Waveform-Preserving Processing Flow of Multichannel Seismic Reflection Data for Adjoint-State Full-Waveform Inversion of Ocean Thermohaline Structure

This paper presents a specific data processing flow to be applied to marine multichannel seismic reflection data collected by a streamer in order to use them to perform prestack adjoint waveform inversion of ocean's thermohaline properties. The overall goal is to increase the signal-to-noise ratio (SNR) of the weak reflections generated at the small impedance contrasts within the water layer while preserving the direct wave. The processing flow focuses on increasing the SNR of the shot gather records by forcing noise amplitudes to fall inside a range of physical plausible values for water layer reflections. This processing step is applied in two independent branches of the workflow; one dealing with the water layer reflections and the second with the direct wave, which are separated by applying a singular value decomposition. To test the performance of the processing flow, we combine actual noise field recordings with a synthetic seismic data set. We apply the proposed data processing flow to quantify differences between noise-free and processed record sections, and we then compare with the results obtained by applying a Butterworth filter (BF). For offsets smaller than 1500 m, the BF processing produces a signal with SNR <; 0.1; the proposed workflow allows to retrieve the seismic signal with 0.1 <; SNR <; 2.4. For offsets larger than 1500 m, the BF processing allows obtaining the SNR up to 1.4, while the proposed workflow increases the SNR up to 5. We finally demonstrate that the processed data can be used to perform waveform inversion with an accuracy of
_~
0.02 m/s.