For abstract, define only those abbreviations that you will use in the abstract, not those you will use in the main text. For example:
Abstract: The sea-surface temperature (SST) in the eastern Pacific Ocean becomes abnormally warm every 3-5 years, known as the El Nino. The authors use SST data derived by satellite to show another El Nino is on the way.
Note that the purpose for defining abbreviations is for speedy writing and reading. If the abbreviation is used only 2-3 times, do not introduce it (unless it saves you a lot of space, say in a long abstract). Avoid using confusing abbreviations, such as AS (just-in-time definition for Arabian Sea), NC (North China), etc. Just-in-time definition is what I prefer; however, most journals in my fields (oceanography and meteorology) still do not accept just-in-time definition. (It’s time for “evolution.”)
For the paper’s main text (not including the abstract), whatever abbreviations you define in the abstract do NOT count. You need to treat the paper as a separated body from the abstract. Do not define the same abbreviation again and again in the paper, especially not in the conclusion part (no matter how convenient this is for people like me who may read the conclusion first or only read the conclusion).
If I am wrong on these rules, please let me know.
p.s. Yahoo! has a nice article on "When and how to introduce abbreviations" under