Coastal Digital Twin: Learning a Fast and Physics-informed Surrogate Model for Coastal Floods Via Neural Operators
Abstract
Coastal flooding is considered one of the most significant impacts of rising sea levels, potentially threatening lives and damaging infrastructure. Increasingly into the future, coastal flooding effects on society will be exacerbated due to the density of coastal populations and the accelerating rate and severity of extreme climate events. One way to predict coastal flooding is to perform simulations using physics-based numerical models, such as Nucleus for European Modelling of the Ocean (NEMO). With a range of coastal regions and physics processes considered, these physical models are mainly driven by wind speed and sea level pressure and simulate the dynamics of water velocity and sea surface height by solving the mass conservation and the momentum equations. Nevertheless, running these physics-based models can be extremely computationally expensive due to the need to numerically solve the partial differential equations (PDEs) at each time step and grid point. Machine learning (ML) models have received much attention in the Earth Science community due to their success at providing fast, data-driven simulation without substantial loss of accuracy. In this work, we present an ML model to develop the first coastal digital twin with state-of-art physics-informed machine learning techniques. For this purpose, we implement the Fourier Neural Operator (FNO), a neural network that can approximate solutions to PDEs 1-3 orders of magnitude faster than traditional PDE solvers. Experiments with our modified Fourier Neural Operator show reliable emulation of NEMO at 100x acceleration. These results show a promising approach to massively accelerate Earth systems simulators, which can enable scientists to efficiently iterate over hypotheses and experiments, and execute many simulations for decision-making and uncertainty quantification.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.A14C..01J