I'm experiencing what appears to be some model instability at the interface of my 2D and 1D domains. I'm connecting an upstream 2D flow area to a downstream 1D reach because I have four bridges downstream. I've messed around with warmup, initial conditions, time step, increasing 1D/2D iterations, etc. but I'm getting sharp increases in velocity and a jump in WSE at the bounding XS. Has anyone troubleshooted a similar issue?
We have run into this on several models. I've played with all the things you played with and also switching between Full Momentum and Diffusion Wave. I have found that the DW runs have much less of these types of issues while FM runs tend to have some strange velocity and WSE issues along any interface. I see in your cross section you have shallow flow on a terrace which is above the flow in the channel. I have seen this phenomena occur with the use of the FM equations in channelized flow conditions.
We have had several models with fully channelized/contained flow that the 2D FM equations over-estimate water surface elevations and/or place water where a 1D model would not. That is to say, if you model the reach with a 1D model you will get a certain answer. Then if you model it with a 2D model with the FM equation set, the water surface will some times be 2-3 feet higher. So what may be happening in your situation is the 2D model is simply giving an answer that disagrees with the 1D model at this interface.
I have not yet found a great answer for this issue. You can try to move the connection to a more favorable location and adjust some roughness values. I have not found much benefit in changing the grid size at the interface. Sometime the warmup will help, but not always. I suppose that as long as the model is not having a volume conservation issue that maybe the answer is to live with the problem and address it in the mapping.
Thank you for the response and advice, very appreciative!
I've only been running 2D in DW and not FM (I've been hesitant to try FM in light of the issues I'm having ).
I tried moving the d/s end of my 2D area to the next 1D XS downstream but interestingly the problem propagated itself downstream - look at that velocity spike!
I've considered scrapping the 1D domain and extending the 2D area downstream (WSE doesn't reach the bridge decks in the events I'm simulating); but, it could take quite an effort to get the abutments and piers accurately represented in the terrain...
Hi, I have experimented with max 1D/2D iterations and have set it as high as 20 but the problem still persists.
Some other tests I have tried:
- Reduce grid cell size at 1D/2D interface (gradually decreased from 25'x25' to 10'x10' at interface)
- Decrease computation time step (tested time steps ranging from 5 sec down to 1 sec)
- Added warm up period
- Modified Manning's n at interface
- Moved 1D/2D interface to next downstream XS in 1D reach
- Tested different initial conditions (use of .rst file versus initial flow versus initial conditions ramp up)
I have not yet experimented with time slicing, and have only run Diffusion Wave, not Full Momentum.
What happens when you vary the minimum flow tolerance? It looks like the water surface is a higher elevation on the 2D side of the interface. Do the water surfaces converge during 1D/2D iterations, or are there 1D/2D flow errors? How large, compared to the minimum flow tolerance?
What do maps of velocity and Courant number reveal? The transition from 2D to 1D should be uniform. How well do the cross section, Manning number, and slope at the interface match upstream 2D channel characteristics?
How persistent is the problem through the duration of the run?
Full momentum can help identify problematic cells. Give it a try and pinpoint those cells. Keep in mind there may be superelevation at the outside of a bend.
Hi Scott, these are great suggestions. I have not tried all of these yet, but will.
I am currently using v5.0.7 defaults for water surface and volume tolerance. Maximum xsec wsel is 0.054 for my latest run; this occurred two XS's downstream of my 1D/2D interface. At the interface, terrain and Manning's n are identical.
I am currently running steady flows to troubleshoot this issue, so the problem is persistent throughout the run (currently 36 hrs at 1 sec timestep). I am using a 12 hour initial conditions time and a ramp up fraction of 0.7.
I will try FM next to see how the model behaves. Here are maps of velocity and Courant. Thanks again for your response.