The short answer is - however you like. The long answer is that it will vary between sensors, and different ones will be better for different purposes. Cryosat, for example, will be working from the radar return from its field of view, which has a resolution of about 250m. I suspect the return profile will indicate the modal thickness per pixel (when everyone in a crowd shouts, democracy rules) rather than the max, min or median. However, when they average together multiple readings, I would expect them to take the median of those readings, or some other form of average which is stable to individual outliers.
For the sonar buoy, it's effectively a point measurement so the question's void - the fact remains that it's the best estimate we have for the thickness at the webcam, given that it'll be measuring the thickness of a point at most a few feet away from the camera.