Without taking a look at the model, my first guess would be that the 1D reach is getting dewatered where it drains into the downstream 2D domain. Since the discharge from 1D to 2D would be at the high end of the 2D domain, the last cross-sections of the 1D reach may be draining very quickly into an empty channel, or lower water surface elevation. Set the output interval match the computational timestep (or to its minimum), and see if it runs even briefly. The downstream end of the 1D reach can dewater in the first time step, causing instability. Adjust the timestep, and consider adding cross-sections.