Use literals in double precision

[tenomoto]

Unlike Python, the type of a numeric literal is implicitly assumed. Float literals are assumed to be real*4. We need to either explicitly add a type like tref = 288.0_p or promote all floats without type specification with a command line option -fdefault-real-8. With tref = 288.0, tref is declared to be real(p), but an imprecise value of real*4 is assigned without _p after the literal or without the option.

[tenomoto]

The difference of surface pressure with and without -fdefault-real-8 (modified minus original).