Physics-informed and machine learning-enabled retrieval of ocean current speed from flow noisea).

This article focuses on inferring deep-water near-bottom ocean current speeds from passive acoustic measurements using moored autonomous acoustic noise recorders (MANRs) deployed over the Atlantis II Seamounts in the Northwest Atlantic. It demonstrates a strong correlation between low-frequency flow noise (below 20 Hz) recorded by hydrophones and current speed, enabling the development of a regression tree (RT) machine learning model trained on data from a MANR equipped with both a hydrophone and current meter. This RT model successfully infers current speeds at a nearby MANR site with only hydrophone data, distinguishing flow noise from ambient sounds such as shipping noise by analyzing spectral characteristics. The inferred current speeds, including extreme events exceeding 100 cm/s at depths over 2500 m, were validated by comparing flow noise amplitude spectra between sites, showing improved accuracy over previous neural network approaches. The study highlights the potential of converting hydrophones into current speed meters to enhance long-term monitoring of deep-sea currents, which are important for benthic ecosystems and sediment transport.