If AGU conferences are anything like conferences in the computer sciences, poster presentations are reserved for work that is more speculative or less rigorous or less compelling, but even by diminished standards its difficult to understand how such poorly supported ideas could make it into AGU.
Acceptance of posters is usually done in the complete absence of anything other than an abstract to evaluate them on in many cases. If the abstract sounds uncontroversial or the poster diverges significantly from the content of the abstract there isn't much the organisers can do about it.
In many cases people are expected to write abstracts several months in advance, I had to write the abstract for my 1st poster at a major conference before I had actually done any of the experiments for it.
I would suspect that for many conferences poster submissions are hardly ever rejected, the constraints of space for posters being much less than those of time for speakers, but I don't know that for a fact.
TTFN,
WK