Is it possible for a scientific article to lose citations?

Google Scholar's citation counts are based on the number of citations found by Google's web-crawler. They tend to fluctuate up and down a little. This is usually either because documents such as internal reports or theses disappear from their authors' websites, or because Google's algorithms identify certain documents as duplicates of one another and 'merges' them.

In addition to the inherent curation problems of web-crawling, determining duplicates, etc, it is quite possible for a document to truly lose citations.

For this to happen, a citing article would be withdrawn or amended such that the citation is removed. If the article is withdrawn for error, one might argue whether it ought to still count as a citation or not, since the article may leave some trace. In the case of, say, entirely fraudulent publications being used to pump citation counts, however, those citations would certainly be well and truly gone.

To directly answer the question:

Is it possible for a scientific article to lose citations?

Not except in rare cases. For the most part, citation is a permanent thing. This also depends on how citations are counted: if an article is published but then updated to a new version (say on a preprint site) which removes a citation, does the previous version count as a citation or not? Citation databases like Google scholar have to make a decision about such matters, and whatever they choose, they try to apply it in some consistent matter. But it can be hard especially as article titles can change between revisions.

Because it is fully automated, Google scholar is likely to be less stable than the "real" citation count as measured by some objective criteria (making a decision as to whether, e.g. preprints or website pdfs are counted, whether updating or retracting a paper removes old citations, and so on). I would guess that the "real" citation count almost always never goes down.