I would guess by normalizing to either a conveniently placed Lambertian reflector - or something close like bare soil or concrete.   Or a high end light meter - or both. 

 I had a hand in developing Deere's HarvestLab calibrations.    NIR and other spectra are fairly mature (albeit purely empiric) research domains.   You have to have standard reflector ( and/or emitters) for this to work over time and space.   Sometimes that means referencing objects in nature.  Lakes, bare soil (ie Lambertian reflectors), etc.  It is otherwise a purely relative data layer that admittedly has some value. 

What is probably not true is a presumption that outside spectra had loads of additional data that cannot be obtained through most of the visible spectrum (and near visible bands).   The whole reason we all evolved to view these spectrum is... that is where the signal is!     We should strive to have better tools (better multispectral analytic tools) within the visible domain ahead of pursuing much else beyond what a CMOS (or CCD) with standard masks can take in.   Plenty of hay to be made still. 

I do  eagerly await the days of cost effective, made to order, multi/hyper spectral cameras in the 10 MP+ range that I can afford to crash into trees.