You may have heard of Oregon’s Willamette Valley because of its agricultural success, particularly when it comes to wine. This is at least partially due to a series of prehistoric flood events that occurred between 10 and 40,000 years ago, carrying tens or hundreds of feet of silt to the valley floor. Kleinschmidt Associates’ Chris Goodell, principal consultant for hydraulics and hydrology, decided to model the last of these events using HEC-RAS, the modelling software developed by the US Army Corps of Engineers for hydraulic simulations.
To his knowledge, the Glacial Lake Missoula floods were the largest magnitude dam breach event ever to be modelled in HEC-RAS. It would test the software’s limits. It would also be a test for Hypernet Labs’ Galileo, an application designed to help engineers get easy access to compute power, which allowed Goodell to run the simulation remotely in order to optimise run time and free up his local machine.
How to set up the HEC-RAS model
Initial inspiration for the model came from Goodell’s discovery of a book by David Alt called Glacial Lake Missoula and its Humongous Floods (2001). The book also provided much of the information necessary to build the simulation. Luckily, much of the flooding occurred in eastern Washington, which is very dry. The geologic evidence isn’t hidden by vegetation, and there hasn’t been a lot of erosion. As a result, there’s a lot of available data. It’s possible to see the scarring resulting from these floods even just by looking at Google Earth.
Goodell was originally working in 1D since he initiated the project in 2008, before HEC-RAS 2D was available. To give some idea of scale, the model originally included 592 cross sections with spacings from 1 to 25 km (with a 5 km average). As he refined the model, there were eventually 2,346 cross sections, 68 reaches, 34 junctions, and 36 external boundaries. The results provided further indication of the outsized proportions of the event, especially with regard to the magnitude of discharge, flow velocities and inundation extents.
What are the advantages of a 2D model?
Significant differences between Goodell’s original 1D and his later 2D results include different areas that are wet and dry. HEC-RAS 2D allows for different water surfaces across any given transect, while the 1D version allows for only one surface across each cross section. The 2D mesh is also much denser than the cross sections knitted together in the 1D version, making the 2D results look more sophisticated.
But one of the greatest differences between the 1D and 2D models was time, both in terms of model setup and run time. Generally, 1D modelling entails more setup time, while 2D modelling means longer run times and more computational power. This is where Galileo presents an enormous advantage, as it can help minimise the 2D time commitment by quickly and easily connecting modellers to optimised machines, freeing up their local computers.
To simulate the Missoula flood using HEC-RAS 1D, Goodell spent about a month of evenings and weekends setting up the model before he could even run it. Taking troubleshooting into account, it was a couple of months before he had a working model that was actually running and providing results. Setup for HEC-RAS 2D is much faster because it determines the direction of flow using the terrain behind the digital mesh instead of relying on the manual input of the modeller’s assumptions. Once Goodell entered a digital terrain surface, the 2D model was up and running within a couple of hours.
As expected, though, the 2D version took longer to run - three hours on his local computer compared to 44 minutes for the most refined version of his 1D model. With Galileo, which connected him to a remote 32 core machine, the 2D runtime was down to only two hours.
How to handle the computational intensity of HEC-RAS
According to Goodell, 1D will continue to be useful, largely because 2D is so computationally intensive: “If you want to model a long reach, it’s just way too prohibitive, and so you can combine 1D elements - for the efficiency - with 2D for accuracy and make a really powerful model that way. A lot of times we’ll use 1D to inform the boundary conditions for a more focused 2D area.”
It’s actually the computational intensity of 2D modelling that makes it such a perfect use case for Galileo software. It helps to remove the barriers to efficiency and make 2D modelling more practical by giving modellers extremely easy access to RAS-optimal machines and allowing them to run their models elsewhere, without tying up their own workstations.
The author is Ivan Ravlich, CEO of Hypernet Labs.