Fortran is very heavily used for development in numerical weather prediction. For example, the WRF model (http://www.mmm.ucar.edu/wrf/users) includes hundreds of thousands of lines of code and has been under continuous development since the late 1990's.
And Fortran has evolved significantly in the last two decades, absorbing many features (and lessons learned) from other languages. Structures, modules, encapsulation, argument checking, pointers, allocatable arrays, classes, inheritance, function overloading, pure functions.... Wikipedia has a decent summary (http://en.wikipedia.org/wiki/Fortran).
I dunno. I haven't looked at what's going on with FORTRAN in a meaningful way in years. It might largely fall on how algorithms designed to exploit long-obsolete vector processors have been adapted to work on modern, highly parallelized supercomputer networks, with powerful but conventional individual nodes separated by really slow networks.
Is no one writing large bodies of new FORTRAN code? Or is it mostly in COBOL style maintenance mode, and e.g. called a lot by Python, which I gather is getting a lot of attention in the scientific community (I started with IBM 1130 punched card FORTRAN "IV" in 1977 in a academic year long class, but stopped after that, it wasn't an answer for the sorts of problems I was interested in, C and Lisp are).