2D HEC-RAS 6.0 Modularization Framework for Implementable Real Time Flood Alert with Surrogate ML
Abstract
Flood damages have devastated Harris County, Texas as well as many other parts of the world over the past decade. In this period, 2D hydrodynamic solvers have improved immensely and could provide more information for real-time emergency management as well as increase the accuracy of flood modeling with integrated control structure operations as previously accomplished with 2D HEC-RAS (Garcia, 2020). These more advanced solvers, however, have an unattainably high computational burden to perform in real time or for optimization modeling without supercomputing or surrogate models which in turn requires even more computational resources to produce a training dataset. The framework introduced here addresses this downside by modularizing the 2D modeling domain, including flood control structures, to minimize the computational needs for updating an ML surrogate long term. Using the geometry and timestep data, a new module of a contiguous subset containing the outlet can be generated from the original domain with identical computation points. Internal time step data from the original runs becomes the external boundary conditions for the module which can then be run in isolation to produce the same results as rerunning the full model with the new domain changes. Running the smaller module to update the surrogate training dataset saves a significant portion of the computation time needed for updates after major construction or other land use changes in the domain and makes the method feasible for long term use at scale. The methodology also maintains a one-to-one mapping of the complex 2D model to the supervised ML surrogate which increases public trust for municipal applications in the future like next generation real-time alert systems, structural optimization, or regional flood risk analysis.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H33D..08G