I changed the method of calculating flows across the lateral structures from 2D equation to weir equation. When I did this I was not able to use the hot start file from the last run (which used 2D equation). The preprocessor said that the hot start file had a breach in the lateral structure plan. I’m not sure why. There were no breaches scheduled anywhere in the geometry when the model run started. This was my only clue as to what had been wrong with the lateral structures.
Changing to the weir equation stopped the leaks that showed in the profile plot. At that point it was a matter of stabilizing the 1D model, which had the tendency to go dry at its head before flow reached it through the 2D area. I got past this by putting an inflow near the head of the 1D model to flood the upstream 2D channel at a cold start. With that in place the only way I found to stabilize the 1D model was to iterate calculation of flow over the lateral structures. It took some testing to choose thresholds for minimum flow and number of iterations. Then it was necessary to find a time step sweet spot between stability of the 2D area and stability of the 1D model.
Overall the model is about 50 percent slower than when using 2D equation for the lateral structure, but it is stable and results look right.