Producing a new in-situ wind speed product for the Southern Ocean based on acoustic meteorology from biologged southern elephant seals

Obtaining in-situ wind speed measurements over the Ocean is paramount to i) better calibrate numerical models and satellite imagery and ii) increase our understanding of upper ocean processes such as Mixed Layer Depth (MLD) dynamics and air–sea gas exchanges. For instance, increasing wind speeds deepen the MLD through surface turbulence and mixing. In this study, we evaluate the use of acoustic meteorology as an alternative to conventional reanalysis and scatterometer products by training regression models and Long Short-Term Memory (LSTM) neural networks on passive acoustic data recorded from biologged elephant seals. The study presents the first LSTM fine-tuned on wind speed data products from the China-France Oceanography SATellite (CFOSAT) mission, exhibiting remarkably low errors with an average error over all deployments of 0.86 m/s . Using an innovative validation protocol, by correlating wind speed predictions with co-located MLD measurement, the different models are compared. All models’ wind speed predictions exhibit better correlation coefficients with the MLD than the ERA5 re-analysis wind speed products, underscoring the added value of acoustic meteorology in resolving ocean-atmosphere coupling. This work highlights the potential of passive acoustic monitoring as a robust, scalable, and accurate tool for oceanographic and climate studies.