Then it looks like underwater landslides, volcanoes and earthquakes could throw things off from time to time.

Traditional dead reckoning is reasonably accurate over moderate timescales in the mostly empty ocean as subs are big and relatively slow. Thus the existing approach of only occasionally surfaces for GPS. This is therefore more a supplement as being able to regularly recenter even a few times a day is good enough.

I would assume that you'd use that data in conjunction with others (sensor fusion / kalman filter).

Wenn you know you were at location X 10 minutes ago and now one sensor tells you you are at Y ... You can reasonably assume that that sensor is wrong.

Doesn't matter just take into account what didn't move. you don't have to recalibrate at each instant.

The earths magnetic field is not constant. I don't know how much it changes, but I know magnetic north drifts a bit every year. And every once in a while the field reverses (IIRC we are like 10k years overdue for a reversal if we read the history of them right - a lot of guesses go into that of cousre)

Right, but it seems it is constant enough in the near term for practical purposes.

I didn't mean to imply that it wasn't useful for practical purposes. That is what the article is about. I just meant to point out that there are limits and disadvantages to work around. I'm expecting to make this useful they will have to have a team to constantly remap the earth, and send those updates to whoever needs the information.

There’s nothing saying that you can’t do a re-localization or remapping path with basically infinite frequency, or any frequency that we know that there is variance around. At a minimum then, it becomes a better standard for which other things can bear on.