2D Shallow-Water Bathymetry Inversion with Physics-Informed Neural Network
Accurate bathymetry inversion can positively impact navigation, coastal planning, and natural hazards protection. Traditional inversion methods are often computationally expensive and rely on extensive, high-quality datasets, making them impractical for large-scale or real-time applications. We explore the use of physics-informed neural network (PINN) constrained by the two-dimensional shallow water equations to estimate bathymetry from sparse and noisy hydrodynamic data. By embedding the governing physical equations into the neural network architecture, our approach ensures physically realistic and accurate bathymetry inversion without the need for large training datasets. The results demonstrate that our PINN-based method can accurately reconstruct bathymetric profiles with limited noisy data. Recent developments in PINNs, combined with advancements in GPU hardware and optimization algorithms, further enable their deployment in real-time applications. This advancement shows significant promise for real-time coastal monitoring and has the potential to enhance predictive models in coastal engineering.
IMPACT
The resulting PINN-based bathymetry inversion framework would provide unprecedented capabilities for hydrodynamic surveying and enable more innovative research in this area. It would directly impact ERDC’s civilian mission by supporting planning for natural hazard prevention, and ERDC’s military mission by enabling rapid wet-gap crossing and joint logistics over the shore (JLOTS) operations. Ongoing improvements in PINN algorithms and graphics processing unit (GPU) hardware will continue to enhance the performance of these tools. The PINN-based framework developed in this effort can be used for problems in any discipline in which the governing equations are well formulated as PDEs.
PRESENTER
Rivera-Casillas, Peter
peter.g.rivera-casillas@erdc.dren.mil
787-647-8651
ERDC-ITL
CO-AUTHOR(S)
Sourav Dutta
sourav.dutta@austin.utexas.edu
CATEGORY
AI/ML for HPC
SECONDARY CATEGORY
Comp Fluid Dynamics
SYSTEM(S) USED
Narwhal