Recent works show that multichannel seismic (MCS) systems are able to provide detailed information on the oceans' fine structure. The aim of this paper is to analyze whether 1‐D full waveform inversion algorithms are suitable to recover the extremely weak acoustic impedance contrasts associated to the oceans' fine structure, as well as their potential to image meso‐scale objects such as meddies. We limited our analysis to synthetic, noise‐free data, in order to identify some methodological issues related to this approach under idealistic conditions (e.g., 1‐D wave propagation, noise‐free data, known source wavelet). We first discuss the influence of the starting model in the context of the multi‐scale strategy that we have implemented. Then we show that it is possible to retrieve not only sound speed but also salinity and temperature contrasts within reasonable bounds from the seismic data using Neural Network relationships trained with regional oceanographic data sets. Potentially, the vertical resolution of the obtained models, which depends on the maximum frequency inverted, is of the order of 5–10 m, whereas the root mean square error of the inverted properties is shown to be ∼0.5 m/s for sound speed, 0.1°C for temperature, and 0.06 for salinity. To conclude this study, we have inverted synthetic data simulated along an oceanographic transect acquired during the EU‐funded Geophysical Oceanography (GO) project. The results demonstrate the applicability of the method for synthetic data, as well as its potential to define oceanographic features along 2‐D transects at full ocean depth with excellent lateral resolution.
Read full abstract