Session: 08-05-01 AI‑Enabled Marine Sensing, Monitoring & Digital Twins

Submission Number: 181140

Physics-Guided Probabilistic Digital Twin for Underwater NoisePrediction 

Ship traffic is a growing source of underwater radiated noise in coastal waters, posing risks to sensitive marine species and motivating the need for operational noise prediction tools. We present a physics-guided probabilistic digital twin for real-time estimation of three-dimensional acoustic transmission loss in complex marine environments. As a representative application, the framework is demonstrated for the Salish Sea along major shipping lanes from the Pacific Ocean to the Port of Vancouver. A database of over 30 million source–receiver pairs was generated using a Gaussian beam solver across seasonally varying sound-speed profiles and one-third-octave frequency bands between 12.5 Hz and 8 kHz. The digital twin integrates four components: (i) a physics-based mean function combining spherical spreading and frequency-dependent absorption; (ii) a convolutional encoder that captures bathymetric variability; (iii) a neural encoder for spatial and frequency features; and (iv) a sparse variational Gaussian-process layer providing calibrated uncertainty. This hybrid learning structure enables robust, data-efficient inference while retaining physical interpretability. The probabilistic predictions support the creation of real-time sound-exposure maps and risk envelopes for ship-noise management. When coupled with near-field source models, the framework allows optimization of vessel speed to minimize acoustic impact while maintaining operational efficiency. The proposed approach demonstrates how physics-guided machine learning can enable uncertainty-aware digital twins for sustainable and intelligent marine operations.

Presenting Author: Rajeev Jaiman University of British Columbia, Vancouver

Presenting Author Biography: Rajeev K. Jaiman is a Professor and Seaspan Chair in Marine Systems Engineering at the University of British Columbia, Vancouver, Canada. He leads the Computational Multiphysics Laboratory, focusing on digital twins, fluid–structure interaction, and physics-guided machine learning for sustainable marine systems. His research spans hydrodynamics, ocean acoustics, and intelligent vessel design, with applications to quiet and energy-efficient ship operations.

Authors:

Indu Kant Deo University of British Columbia
Akash Venkateshwaran University of British Columbia
Rajeev Jaiman University of British Columbia, Vancouver