2025_programme: A Machine Learning Approach for Model Selection in Underwater Acoustic Propagation

• Day: June 16, Monday
    Location / Time: C. THALIA at 15:50-16:10
• Last minutes changes: -
• Session: 09. Modeling techniques for underwater acoustic scattering and propagation (including 3D effects)
  Organiser(s): Boris Katsnelson, Pavel S. Petrov
  Chairperson(s): Boris Katsnelson, Sven Ivansson, Pavel Petrov
• Lecture: A Machine Learning Approach for Model Selection in Underwater Acoustic Propagation
  Paper ID: 2314
  Author(s): Finn Henman, Stewart Haslinger, Joshua Wakefield
  Presenter: Finn Henman
  Abstract: Underwater acoustic propagation models use different assumptions and approximations to simplify the wave equation. These assumptions make different models more or less suitable for modelling propagation in different ocean environments, such as a ray tracing, parabolic equation or normal mode model. Ensuring that one's choice of underwater acoustic propagation model is tailored to a given scenario is important to ensure propagation is modelled as accurately as possible.\n\nCurrent methods for deciding which model to use are often based on simple thresholds, for example, using a ray tracing model for simulations with a wave frequency above 1000 Hz and a parabolic equation model below 1000 Hz. A tool that selects the most suitable model based on multiple environmental parameters would be immensely useful for a variety of underwater acoustics applications. This paper presents a novel method for selecting the most appropriate model for a given scenario by using a random forest classifier to predict the best model based on the input parameters.\n\nA training set of ocean environments was generated using a variety of depths, ranges, bathymetry and sound speed profiles.\nThese environments were passed through the RAMSurf (Parabolic equation) and BellhopCXX (Ray tracing) propagation models and the results were compared to Kraken (Normal mode), which acted as an arbiter. This provides a model selection backed by empirical data, which takes into account a wider selection of environmental parameters, allowing for edge cases to be modelled more accurately. The results show a high level of accuracy, with the potential to be further expanded to include a wider range of propagation models and for alternatives to the selection metric.
    Download the full paper
  This paper is a candidate for the "Prof. Leif Bjørnø Best Student Paper Award (for students under 35)"
• Corresponding author: Mr Finn Henman
  Affiliation: The University of Liverpool
  Country: United Kingdom