A while back Peter S asked for authoritative opinion on whether a boundary
defined by a river would move when the river course shifted, or stay along
the historical track. I found one example of a boundary that did move with
the river - the US/Mexico border on the Rio Grande - but couldn't say if this
was a feature of a general principle or a specific agreement.
Now I think I've found the general authoritative statement. Brownlie, in his
introduction to African Boundaries, has a section call Some Technical
Problems. In it he states: "A related question is whether changes in the
breadth or course of the river involve changes in the boundary. The
presumption in the case of reference to the line of a river, or its thalweg,
is that the boundary follows the feature."

Grant