*albiz from *albʰós
Other IP user here. I would usually not touch this with a nine foot pole as they say, but your wording "must" is too strong, even if it can't be regularly derived directly from *albʰós, f. nom. *albʰéh₂, f. loc. *albʰéh₂(i), etc. that I have to believe you, if you meant that it couldn't be inherited.
The PGem root is "river", but Kroonen says only that it's probably from the river "Elbe" (one of the major confluences in central north Germany).
I guess that it is possible the inflection was innovated after PIE. This might be due to "Elb'" being the combining form (Elbgewässer), or any other reason. "derived" can well come with a breaking change. The status quo is being not that precise about the inheritance template anyway, are we? Maybe we should be.
The association to "white" is reasonable, reflected in the literature, even if the typology is perhaps naive. The problem is, I expect, to find non-German comparands that match exactly, which would require specialized literature on hydronymy to check.
It may be best left unrooted for the time being.
A counter opinion for OP to consider, *albʰos- alternatively gives \*h₂elbʰos-, because \*h₂e, dubbed the a-colouring laryngeal, frequently reflects a in descendents. In contrast, PIE \*a is marginal, even contestable. Anyhow, *h₂el- (“to grow, old”) could make sense here. And if not, then \*a may be prefered as a clue to a substrate language, e.g. old-European-Hydronymy, which is perhaps still a little naive. 109.41.1.176 17:23, 8 March 2020 (UTC)