Background Temperature Images of Mesoscale Ocean Features from Laplace and Laplace-Fourier Domain Seismic Waveform Inversion

2-D temperature images of the ocean with resolution within a few tens of meters in distance and depth can be recovered from conventional marine multichannel seismic (MCS; low frequency acoustic) data via full waveform inversion (FWI), as demonstrated by several research groups in recent years. A primary limitation with FWI is that the more computationally efficient local inversion methods require an accurate estimate of the background sound speed in the material as a starting point to avoid converging to a local, rather than global, solution. In the ocean, expendable instruments are often used to obtain 1-D temperature and sound speed profiles; in typical MCS data collection, however, expendables are deployed just once per day, resulting in only one hydrographic profile every few hundred kilometers. In addition, the band-limited nature of seismic data, which typically lacks reliable frequencies below 5 Hz, makes it inherently challenging to extract the long wavelength sound speed directly from seismic data. Laplace domain inversion (LDI) developed by Changsoo Shin and colleagues requires only a simple starting model to produce smooth background sound speed models without requiring prior information about the medium. It works by transforming input data to the Laplace domain and then examining the zero frequency component of the damped wavefield to extract a smooth sound speed model. Laplace-Fourier domain inversion extends the technique to include additional frequencies below 5 Hz. This ability to use frequencies below those effectively propagated by the seismic source is what enables LDI to produce the smooth background trend from the data. We applied LDI to five synthetic data sets based on simplified models of oceanographic features and recovered smoothed versions of our synthetic models, demonstrating the viability of this method for creating sound speed profiles suitable for use as starting models for other FWI methods that produce more detailed models.