Surrogate Model for Shallow Water Equations Solvers with Deep Learning

Abstract

A surrogate model for shallow water equations solver was developed using a point-to-point prediction approach.• The new model overcomes the limitations in raster-image based approaches.• The model can successfully predict flow fields and respect physical laws with high accuracy.

Modeling the Flow and Geomorphic Heterogeneity Induced by Salt Marsh Vegetation Patches Based on Convolutional Neural Network UNet‐Flow

“…Based on the previous research in the field of flow field reconstruction (Guo et al, 2016;Ribeiro et al, 2021;Song et al, 2021), we introduce the UNet-Flow neural network architecture, which is an enhanced version of the wellknown U-Net model. U-Net is a widely used CNN architecture proposed by Ronneberger et al (2015), typically employed in image segmentation tasks, including medical image analysis.…”

“…However, this approach can significantly impact accuracy. In recent years, with the rapid development of deep learning, deep neural networks have been attempted to solve computational fluid dynamics problems (Guo et al, 2016;Ribeiro et al, 2021;Song et al, 2021). Data-driven machine learning methods can generate accurate approximations of simulation results using limited computing resources, known as the "surrogate model method" (Song et al, 2021).…”

“…In recent years, with the rapid development of deep learning, deep neural networks have been attempted to solve computational fluid dynamics problems (Guo et al, 2016;Ribeiro et al, 2021;Song et al, 2021). Data-driven machine learning methods can generate accurate approximations of simulation results using limited computing resources, known as the "surrogate model method" (Song et al, 2021). This method has been preliminarily applied to reconstruct steady-state aerodynamic flow fields (Bhatnagar et al, 2019), flow field prediction over airfoils (Hui et al, 2020;Sekar et al, 2019) and vehicles (Guo et al, 2016).…”

Bathymetry Inversion using a Deep-Learning-Based Surrogate for Shallow Water Equations Solvers

“…Figure 1 shows the architecture of the CNNbased surrogate model developed in this work. Similar structure has been used in Guo et al (2016) for steady-state, laminar Naiver-Stokes equations, and Forghani et al (2021) and Song et al (2021) for 2D SWEs. The main differences between our surrogate model and that in Forghani et al (2021) (named AE in their work) are in the detailed architecture.…”

“…Therefore, there is a trade-off between the accuracies of forward prediction and inversion using CNN-based surrogate models. It is beyond the scope of this work to investigate whether we can and if so, how to in Song et al (2021).…”