IAHR World Congress, 2019

21st Century Technology to Support Flooding Solutions: Use of LiDAR, UAEs, and High Performance 2D Numerical Flood Simulations using Cloud Computing

Reinaldo Garcia 1 Jacinto Artigas 1 Javier Fernandez-Pato 2 Mario Morales-Hernandez 2 Pilar Garcia-Navarro 2 Nicholas Calero 1
1Applications, Hydronia LLC, USA
2Fluid Mechanics, University of Zaragoza, Spain

LiDAR (light detection and ranging) and UAEs (Unmanned aerial vehicles) technology are making available huge amount of high resolution elevation data for hydrologic and flood simulation models. However, the application of two-dimensional (2D) models that make use of this data confronts several obstacles, including the model limitation to handle the enormous quantity of elevation points, and very long runtimes, which often make the use of 2D models impractical if not impossible. There have been attempts of square cell models to large scale flood applications, but this type of model forces using very small cell size all over the modeling area, leading to excessive number of cell elements. The development of parallelized versions of 2D numerical finite-volume algorithm for Graphic Processing Units (GPUs) on flexible meshes, is radically changing the hydrologic and hydraulic modeling practice. However, until recently, the cost of top-of the line hardware had prevented widespread use of this technology. In this work we discuss large scale applications of the RiverFlow2D model running in the Google Cloud to simulate flooding forecasts along a 420-mile reach of the Red River of the North, USA.The goal was to have a single integrated hydraulic model rather than separate models for the long reach while providing enough resolution to realistically represent the flooding of large areas. One to 4 million cells were generated using a plugin developed for the QGIS open source Geographic Information System. The model was calibrated with high-water marks from historic flood data. In order to perform simulations remotely and using the most advanced hardware, the model was implemented in Google Cloud VMs (Virtual Machines) with NVIDIA Tesla V100 cards. The GPU Cloud model allowed reducing runtimes from 50 to 600 times costing less than 2 US$ per hour. Results demonstrate that large scale applications, that were not feasible until recently can be realistically done using the GPU flexible mesh models in the Cloud.

Reinaldo Garcia
Reinaldo Garcia








Powered by Eventact EMS