Benchmarking Models for Ocean Wind Speed Estimation Based on Passive Acoustic Monitoring

Passive acoustic monitoring is a promising tool for long-term ocean observations, offering a unique means to capture physical and biological processes. This study explores its potential as a source of fine temporal scale in-situ wind speed product by assembling a unique corpus of acoustic datasets co-located with or near in-situ weather stations. This study offers two key contributions: i) setting up a benchmarking framework for the development and evaluation of models in acoustic meteorology, and ii) applying this framework to assess the performance of various models, comparing parameters from the literature with those trained on datasets from this study's corpus. Regarding the latter point, results show that most untrained models fail to generalize due to the intrinsic variability of soundscape in different basins and environmental conditions, as well as calibration inaccuracies. However, all models can achieve satisfactory performance on specific datasets after training. Incorporating diverse observational sources, such as gliders and BGC-Argo floats, could enhance model robustness, and improved acoustic-based estimates will help refine satellite-derived wind products and numerical weather predictions, ultimately advancing global wind field modeling and air-sea interaction research.