Don't get me wrong: python probably is the main language used in research. However there's software that needs to be fast at crunching numbers, I work in computational chemistry and pretty much any reliable software is either Fortran or C++. Indeed you have python libraries, but most are just wrappers.

You have Gaussian: https://en.wikipedia.org/wiki/Gaussian_%28software%29 GAMESS: https://en.wikipedia.org/wiki/GAMESS_%28US%29 CP2K: https://en.wikipedia.org/wiki/CP2K Mopac: https://en.wikipedia.org/wiki/MOPAC

Now, most people do not work in Fortran, but it is something you learn a little bit when you start working in computational chemistry. It happens sometimes to have to debug a software not working or to have to write a module to test an hypothesis. People writing those softwares are also researchers, but mostly are full time dedicated to the software. Generally, there is a huge lack on investment on the software infrastructure, very few people are dedicated at maintaining software that is used by hundreds of thousands of people.

While hiring people, I am satisfied as long as they know a bit of python, but knowledge of Fortran really stands out and highlights a more thorough education. If I have time, I do give all the people an introduction to Fortran, as it is still something you often come across in our field. But yes, unless you're working on the development of such software suites, Fortran is not that common now. You'd publish a proof of concept in python or Julia and then wait for someone else to implement it in one of those libraries.

Someone else using fortran in research checking in. In particle physics, were basically writing huge, physics heavy Markov chain monte Carlos in it. Just one example.