Stable unsteady stream 1D model goes unstable when defining a trapezoidal cross section downstream
I am currently analyzing a channel design for a stream currently characterized by irregular cross sections and an inaproppiately sized second culvert, and I'm making use of an unsteady model for this purpose.
Modeling the current situation of the stream and reaching a stable result was relatively simple, which surprised me considering the irregularities of the channel. I considered a computation interval of 10 seconds, with a flow hydrograph that takes 130 minutes to reach max flow. Results here.
However, modeling the stream with a trapezoidal design from the first culvert onwards (replacing the second culvert in the process) while keeping the same flow parameters has given me really unstable results, as shown here.
After several iterations, including different computation intervals and increasing my lowest initial flows and hydrograph flows, the best results I've gotten have come from interpolating XSs upstream of the first culvert (the area that it's not getting the trapezoidal design). These lead to seemingly stable results downstream the first culvert, but unrealisitic results upstream, as shown here.
Seeing the model get so unstable after giving it a more regular section downstream caught me off guard. I'd appreciate outside opinions on what may be causing this issue and any possible ways to troubleshoot it.