Just be careful: PCI and related procedures are useful to search for possible relations but they are extremely difficult to use.  For example to my knowledge you cannot go to a plot of land and tell a priori where you will be on a PCI set of axes.  I have found gradients analysis (a la WHittaker) backed by an understanding of bioenergetics   to be the way  (e.g.  Hall, C.A.S., J.A. Stanford and R. Hauer. 1992. The distribution and abundance of organisms as a consequence of energy balances along multiple environmental gradients. Oikos 65: 377-390.).   It also allows you to plug into computational analysis much better, through predictions of landscape drivers such as temperature and moisture.   Of course PCI can help you to understand what relations are most important to begin with.  Maybe.  

Firstly, you could approach your Q empirically by using both the NDVI and the PCA-NDVI in your SDM.

Secondly, you could develop hypotheses on explanatory environmental variables in the SDM, based on what is known about the plant(s); NDVI (green biomass) could predict cover types, hyper-temporal NDVI (MODIS or SPOT) provides an estimate of vegetation phenology and may predict ecosystems (combined climate, substrate and cover type mainly) and often delivers a better local climate prediction than meteostation extrapolations.

Attached an article on use of NDVI in plant SDM and development of hypotheses on explanatory variables in SDM.   

Article Fine resolution distribution modelling of endemics in Majell...

The second components is The change detection Image. Specially, Selective PCA extract de parameters of areas with small amount of changes.

Nothing like good ground truth form a skilled botanist.....both unto itself and to calibrate what ever you are doing.    Also its good to revisit the Clements-Gleason debate, and Whitaker's empirical support in Smokey Mtns for Gleason's individualistic concept.