You are not missing something. It would be more accurate to say “ultrasonic stage height sensors”, because depth is merely retrieved, not sensed. This is problematic when one cannot easily survey the sensor as you said, even more problematic for sensing streams with moving beds. Presently, something like a pressure transducer can measure depth in-situ, but non-contact methods almost universally require an a priori stage-depth relationship. In the early 2000s, the USGS (John Costa, Kurt Spicer, et. al.)showed it’s possible to measure depth with ground-penetrating radar, but that’s not exactly economical or practical.
One of my current projects is tacking exactly that problem. It’s not ready for prime time yet, but if you are working on similar technology and wish to collaborate, feel free to private message me.