Hi Ben,
Yes, that seems to be due to instabilities. If you get warings in the log file indicating that bed levels are different at rivers' junctions, then you should indeed try to correct these discrepancies of bed levels. To do that, you can indeed add cross sections at river junctions, and then adjust their levels.
You can also look at the results at the first time steps: if there is a big change in the results between the initial conditions and the following time steps, it may also be cause of instabilities. In that case, this has to be avoided by improving the initial conditions.
There are other things that can help to improve the stability, but an important one is the time step: depending on the spatial discretisation and the geometry of the river, you may need to decrease the time step to e.g. 1 s. So you can also consider reducing this time step, for example if you kept the default value.
Best regards,
Mathieu