Flood modelling in the Assiniboine River Basin with automated mesh generation using 2D HEC-RAS

Floods are one of the costliest types of natural disasters. Flood simulation models play a critical role in flood risk prediction and damage prevention by delineating areas at risk of flooding and aiding the design of flood protection infrastructure. 2D hydrodynamic models can be used to simulate floods with high accuracy in complex topography or when detailed hydraulic outputs are required. These models are typically composed of terrain data, a computational mesh, arcs and polygons that form the mesh structure and affect cell size and orientation, boundary conditions, and a set of numerical equations representing flow dynamics. 2D hydrodynamic models can be time-consuming to configure, specific model generation steps can be subjective and based on modeller judgement, and they can be challenging to reproduce. This research focuses on increasing the accessibility of critical flood risk information by creating an automated workflow to generate hydrodynamic models in 2D HEC-RAS.

Models will be generated in the Assiniboine River in Manitoba, Canada. This river is bordered by urban areas and flat prairie topography, which present unique hydraulic modelling challenges. Models will be developed in test sites with varying topography to ensure the generalizability of this work. The 2D hydrodynamic software 2D HEC-RAS will be used, as it is publicly available and widely used across North America.

The workflow includes managing GIS data, generating the computational mesh, and adjusting computational parameters based on desired runtimes, model purpose, desired accuracy, and characteristics of the computational mesh. Automatic mesh generation is already an active area of research; however, the selection of mesh cell size is seldom well-justified, and only semi-automated approaches have been implemented in 2D HEC-RAS. The mesh is designed based on terrain characteristics, flow characteristics, numerical stability, and model accuracy and efficiency. Models will be developed at each test site using the automated workflow to generate an unstructured mesh, a manually generated unstructured mesh, and a manually generated structured orthogonal mesh with a consistent cell size. Each model will be compared in terms of accuracy and computational effort, and qualitatively in terms of mesh configuration. Given that the comparison between the manually and automatically generated unstructured meshes will depend on the modeller, a workflow will be created for the manually generated unstructured grid to increase transparency and reproducibility and to highlight the subjective steps involved in manually developing a model mesh. Both automated and manual workflows will be designed to ensure models and data are findable, accessible, interoperable, and reusable (FAIR principles).

This research will decrease the time required to develop 2D hydrodynamic models for applications such as flood mapping, producing training and validation data for machine learning models, water quality and sediment transport analyses, and stream crossing, control structure and flood protection infrastructure design. With reduced model development time, more time can be spent on model analysis. In addition, this work will increase model reproducibility, enabling more efficient uncertainty and sensitivity analyses, greater transparency in scientific experiments, and the repetition or expansion of experiments in other geographic locations or under different flood conditions.