There are two types of ARGB true colour.
One uses a "float" in the raster
the second type of ARGB use 0-255 inclusive to set the opacity.
By completeness that's four(different values of the "static field constant" for BufferedImage type) types of true colour Alpha image.
"Pre" means one byte of alpha information is there for the whole/complete set of pixels.
so:
pre with "float(one??? 8 bit byte)" or "float
pre with "int(one??? 8 bit byte)" or "int
*

i'm not sure how the single value for the opacity(alpha value) setting is stored in the header or inside the runtime engine under the common image spec(doesn't actually matter).
An alpha raster you can obtain from a buffered image.
AlphaComposite class is mostly for getting two images and combining the pixels(blending) by a set of rules as stated in the API docs.