We went through the creation of a curve-linear mesh in class. We also covered some tips and rules of thumbs in class for specification of boundary and initial conditions. The help is pretty instructive in this topic.
Based on the computational meshes we built in lab, and the sloppy boundary conditions and initial conditions we all specified, many of us got unrealistic results and some did not even get models that converged.
Now you need to go back through more carefully and clean things up to get a reasonable solution you can verify. Some of the things you should think about:
- Quality and resolution of your computational mesh.
- Accuracy of location of upstream and downstream computational boundaries relative to actual location (i.e. you may want to bring in some Auxillary points to more accurately locate these boundaries)
- Accuracy of boundary and initial conditions you specified
- Specification of roughness boundary condition. Was the value we used at all reasonable? Work with the data you have.
- Double check the flow boundary conditions... was Q calculated correctly? How about that WSE?
- If you're having trouble getting the model to converge, see some of Susannah's notes below.
Once you do get a model that looks reasonable, how are you going to verify it? There are the Verification tools, but you will need to enter in both the water's edge points and the velocity points to verify these (as *.anc files). You can also export out the data to ArcGIS and compare your water depth map and velocity maps with the spot measurements.
You may find in verifying your results that things are not reasonable, you may be over predicting velocity and under predicting depths. Think about what boundary conditions and/or intial conditions could be causing or influencing this. Can they be 'calibrated' adjusted until your water depths and velocities match better?
After verifying and calibrating (if necessary), you could then attempt to use the rating curve information to run the model at a range of steady state discharges.
Notes from Susannah
Below are some excerpts and notes from Susannah Erwin, that she took when she was taking the MDSWMS course from John Nelson and company:
Takes less than a full step in the numerical solution. U- velocity relax., E - WSEL relax., and A = global. Relaxation slowly adds change from one time step to the next. 0.4, 0.2, and 0.2 aredefaults. Balance lowering of relaxation by increasing iterations. As a system steepens, increase the ratio between E and (U & A). U and A should stay equal. For steeper systems like the Trinity, set E smaller than U & A, e.g. 0.3, 0.6, 0.6. Relaxation parameters should stay in the tenths range. For really complex systems (e.g. shallow flow over complex topo) you can go down to the hundreds (ex: 0.05, 0.1, 0.1).
Look to initial conditions and roughness first to ensure they are well defined before playing with the relaxation parameters.
Troubleshooting Models that Crash or Don't Converge
Dealing with crashes: Run to model 1 to 10 iterations prior to crash to find problem. Then look at RMS plot longitudinally and look for potential recirculation at downstream end. If it is crashing in the first time step use the "Number of iterations" to define when to stop. If it is crashing on a subsequent time step, use the "Debug Stop"
Three rules of thumb for why model crashes:
Recirculation at downstream boundary: fixed by using downstream grid extension. Extends last cross section at specified slope. Sensitive to slope – don’t want the water to speed up or slow down i.e. backwater. Pay attention to downstream stage specified. The grid extension option extends bedslope down but not water stage. An alternative way of correcting recirculation at the downstream boundary is to set a downstream velocity boundary. The numbers are meaningless, just use check box. Forces all streamwise velocities at the downstream boundary to have a non-negative value. This method affects the upstream solution by about 2 channel widths upstream. Best to start and stop model in straight uniform reaches.
- Converging topography at upstream boundary forces streamlines into the bank creating instability. Best to set upstream boundary where channel is slightly divergent (versus downstream boundary likes slightly convergent). Alternatively you can set a specified velocity distribution with a non-dimensional velocity, non-dimensional cross-stream station, and an angle. Just don’t want to force those into the bank so the solution can’t recover. Use topographic bump to generate bars instead of inflow angle.
- Lateral eddy viscosity (LEV): don’t damp the solution by artificially raising the viscosity. (this is where I suggested that using the variable LEV might help with some scenarios) Rule of thumb: velocity x depth x 0.01; this provides a conservative estimate of LEV