Physics Informed Neural Networks To Model The Hydro-Morphodynamics Of Mangrove Environments

Modelling the hydro-morphodynamics of mangrove environments is key for implementing successful protection and restoration projects in a climatically vulnerable region. Nevertheless, simulation of such dynamics is faced with computational and time constraints, given the nonlinear and complex nature of the problem, which could become a bottleneck for large-scale applications. The recent advances in machine learning, specifically, in physics-informed neural networks (PINNs), have gained much attention due to the potential to provide fast and accurate results, while preserving the binding physics laws and requiring small amounts of data in contrast to other neural networks. In this sense, such networks encode the physics equations into the neural network, and the latter must fit the noisy observed data whilst minimising the equation residual. This study investigates the application of PINNs to quantify the capacity of mangrove environments to attenuate waves and prevent erosion, and represents the first application of PINNs to model vegetation for a large-scale geographical domain with complex boundary conditions. Navier–Stokes, the broadly used mathematical equation to solve for fluid dynamics, is used as the governing equation that constrains the neural network to respect the conservation of mass, energy, and momentum. The Sundarbans, the largest mangrove forest in the world located between India and Bangladesh, is taken as a case study. The results demonstrate that the developed model is superior when compared to a numerical finite element model, in terms of time and data efficiency, yet produces equally strong overall results.