Tuesday, October 8, 2013

Stabilizing a Dynamic Unsteady HEC-RAS Model

Written by Chris Goodell
Copyright © 2013.  All rights reserved.

One of the most frustrating aspects of unsteady HEC-RAS modeling can be the model stabilization process. You know, you’ve gone to great lengths collecting the best survey/topo data and solid hydrology. Then you’ve painstakingly spent hours…possibly days entered all of that data only to find that once you press the “Compute” button, the model crashes. The dreaded “Red Bar”! clip_image002 Sometimes you can get your simulation to complete without crashing, but the listed numerical errors are so high that you can’t with good conscience submit that as your final simulation. Either way, approaching an unsteady HEC-RAS model (especially a dynamic one) as a beginner with little experience and understanding of how to stabilize it can cause significant delays in your project and worse, completely blow up your budget. I’ve uploaded two HEC-RAS projects to the following Google Drive Site:

RawleyResUnstable.prj is an unsteady flow dam breach HEC-RAS project recently sent to me for help. Although the model ran to completion without crashing, it had unacceptably high errors.
RawleyResStable.prj is the fully stabilized version of the model with no numerical errors.

The following lists out the courses of action taken to stabilize the model. Feel free to download the “Unstable” and “Stable” models and try these techniques on your own. The links following some of these items will take you to more information about that particular technique.
1. Cross Section Spacing. The initial spacing was way too coarse. A visual check alone of the geometry schematic and profile plot should encourage you to investigate a finer cross section spacing.
Geometry Schematic
Profile Plot Samuels equation suggests anywhere from 15 ft to 50 ft spacing (depending on what bed slope you use). I interpolated to 50 ft for the entire reach.

  1. Dam Breach models typically have time steps on the order of a minute or less. This model was initially set with a computation interval of 10 minutes which is high even for the largest and “slowest” of dam breach models. I changed the time step to 10 seconds. The selection of 10 seconds was based on “gut” feel and lots of experience doing dam breach models. There are some methods for approximating good timesteps, notably the Courant Condition and Fread’s equation. Also, setting the cross section spacing/timestep ratio equal to a representative stream velocity will get you close. In this case, the cross section spacing is 50 ft, and I was able to pull some velocities (prior to the model crashing) at about 6 ft/s. That suggests a time step of 8.3 seconds. Close enough to 10 seconds, so we’ll stick with that.
  1. I changed the downstream boundary normal depth slope from 0.01 to 0.001. Not sure what is downstream of the first cross section, but 0.01 is awfully steep and was setting up a very low depth at the downstream boundary (which was causing instabilities).
Notice in the following profile plot of the downstream end of the reach how the water surface at the boundary cross section is below critical depth (the red dot). This creates an overestimation of the water depth at the next upstream cross section, which in turns creates some instability over the next several timesteps.
The proper way to handle this would be to find out what is downstream of your model and select a boundary condition that best represents those conditions. In absence of downstream conditions, the 0.001 slope for normal depth provides a reasonable solution.  Either way, this underscores the importance of moving your downstream boundary far away from your area of interest in your study reach.  That way the errors that do originate from your downstream boundary assumption will have diminished to negligible levels before impacting your area of interest. 
  1. In the unsteady flow editor, the initial flow and the first time step flow should always be equal. In the original model, the initial flow is left blank (which is actually okay because RAS will use the first timestep flow if left blank). However, the first time step flow is very low at 0.14 cfs. Peak discharge of your inflow hydrograph is around 600 cfs. I put in a baseflow (minimum flow) of 5% of this, which is 30 cfs. Just to be safe, I put 30 cfs in the initial flow input box as well. It’s important to make sure that the 30 cfs baseflow does not have an impact on the peak of the breach outflow hydrograph.
  1. The upstream end of your reservoir is very steep (14%). This one was easy to spot in the profile plot because of the very nice smooth profile plot followed by a sudden spike in the energy level (green dashed line) at the upstream end. Notice to that there is a corresponding supercritical solution at the upstream most cross section. These are both indications that more cross sections are needed here.
I added more cross sections by interpolating the steep slope at the upstream end of the reservoir (10 ft spacing).
Notice the energy spike is not completely gone, but it is much better. Refining the HTAB parameters (, turning on mixed flow and bumping up n values will collectively take care of the rest of this (see numbers 8, 9, and 10 below).
  1. From RS 4577 to 3791 you have your main channel defined in a small elevated side channel. The right overbank is lower in elevation than the channel.
clip_image020 This is problematic in RAS. Using the graphical cross section editor, I redefined the bank stations to get rid of this problem.

  1. Notice now in the HTAB parameters property table, the HTAB definition begins at the elevation of the invert of the side channel that was previously defined as the main channel. There is a big gap in the computation points in the newly defined main channel. This poses problems with HEC-RAS, particularly at low discharges/stages, since RAS will have to extrapolate to obtain a solution over that range.
I readjusted the cross section HTAB parameters to reflect the new bank station definitions (I used the “Copy Invert” button for all cross sections). While in there, I maximized the resolution of the HTAB parameters by changing number of points to 100 and minimizing the increment as much as possible while still having full coverage of each cross section.
  1. In the figure below, notice all of the “red” areas in this zoomed in section of the  upstream end of the reach.
These are areas that are near critical depth. I turned on the “Mixed Flow” option in the Unsteady Flow Analysis window, which helps stabilize the near critical depths at the upstream end of the reservoir. If you’ve been following along correcting your own copy of the model, you’ll notice that the model now finally runs to completion without crashing. There are still a few minor errors as shown in the Computation Messages.

It’s debatable whether you need to take care of these relatively small errors. Particularly for a dam breach model, where the presumed errors of the input data probably far outweigh these small numerical errors. Nevertheless, there’s a sense of pride in putting out a very robust model, completely free of numerical errors. So…let’s continue.
  1. The errors in the computation window above suggest a problem around river station 6200. That happens to be the upstream end of the reservoir and the source of the error can clearly be seen with the energy grade line and critical depth turned on.
The section at the upstream end of the reservoir is very steep at about 13.5%. In fact, that’s greater than the suggested maximum slope of 10%, as stated in the HEC-RAS manuals. You can see as the pool starts to lower, the steep reach is exposed and because of the low n values, the water surface is calculated to be supercritical. Let’s assume that the geometry is correct, and the upstream end of the reservoir really is that steep. Jarrett’s Equation suggests very high n values should be used here (~0.2), based on the bed slope and hydraulic radius (after the breach when the reservoir has drained). The original model had 0.07 and 0.035 for the overbanks and main channel, respectively. I changed all the n values in this steep reach to 0.15, because I think 0.2 might be a little overkill.
There…no errors! However, there is still a Warning about extrapolating above/beyond the rating curve at a bridge (R.S. 277).
  1. To get rid of that warning, I increased the HTAB headwater maximum elevation at River Station 277 (the bridge) from 278 ft to 280 ft so that RAS doesn’t have to extrapolate.
There we have it. A clean solution. No errors, no warnings.

Thursday, October 3, 2013

Downstream Boundaries

Written by Chris Goodell | WEST Consultants
Copyright © 2013.  All rights reserved.

I just received this question about downstream boundaries in HEC-RAS, and thought it would make a nice post here on

"...regarding the boundary condition, if I use critical depth instead of normal depth will it be unacceptable?  When choosing the normal depth I need to enter a slope, can i use just one slope for the whole stream?"

Either critical or normal depth will work fine IF they are far enough downstream from your area of interest.  Normal depth is usually better, because it is typically a better representation of actual stream conditions.  Critical depth is only accurate when used at a significant grade break, a drop structure, a waterfall, etc., where flow passes through critical depth.  Critical depth does not happen in natural streams very often, and usually the water depth is nowhere near critical depth.  Using critical depth for a typical section of natural river is wrong.   That’s why I tend to always use normal depth.  However, using critical depth as a boundary condition is convenient (it’s easy to use, right?!?!), so it does have some merit. 

For normal depth, the slope you enter is not the slope of your modeled reach.  It represents the slope of the reach downstream of your downstream-most cross section.  You can often times approximate this slope by using a topographic map and locating where topo lines cross the stream, then measure the streamwise distance between them.  But remember this should be done DOWNSTREAM of your downstream cross section.  Normal depth as a boundary condition in RAS is applied to one cross section only-the downstream cross section, and using it means you are assuming that the reach just downstream of your downstream cross section is flowing under uniform (normal) conditions.  This too, is almost never the case in a natural stream due to the constant variation of cross section size and shape.  However, it is much closer to the true solution than critical depth is. 

The bottom line is both methods (in fact ALL methods) for assigning a downstream boundary have some inaccuracies.  That is why it is important to maintain distance between your downstream boundary and the area of your model that you are interested in.  One of the great things about the solution scheme in HEC-RAS is that it is "self-healing".  In other words, if there is an error at a specific location (due to a bad boundary condition, improper n value, poor survey data, that error will diminish the further away from that location you move.  At some point upstream, the effect of that error will no longer be "felt".  How far downstream you need to place the downstream boundary is variable, but you can test this by trying different downstream boundary assumptoins (critical depth and normal depth, or simply normal depth with a bunch of different normal slope assumptions) and seeing how far upstream you need to be to not see an effect from different downstream boundary assumptions.  This post demonstrates this technique: