That’s not a bad question! If it were the case that Fortran was a language that had very little utility outside of a few small things that no one wants to use, the cost of training people would eventually surpass the one-time cost of a rewrite.
As it stands however, Fortran is still a perfectly viable language if you know how to use it, and (one of) the de-facto standard in quite a few environments. So even if you re-wrote the code base, your new guys would still probably have to learn it in order to use some common libraries and tools.
Also, it’s hard to overestimate the complexity in this kind of re-write. We’re talking about a lot of code that is written for performance rather than readability, and where the documentation for the algorithms typically is “that article”.