So, basically, “Ground Truth” is high quality reference data.
Ground truth for mapping projects is databases of precise, on the ground measurements that you can use to match with your IMPRECISE images and maps in order to make yours more precise.
Ground truth genetic information might be the Gene Ontology database.
Ground truth for musical styles is trickier but there are several attempts out there for references. https://www.music-map.com/ is probably the best non-commercial ground truth. Commercial entities like Pandora have their own ground truth data.
Ground truth for object shapes in 2D and 3D is in a number of places, depending on need. https://www.cs.bu.edu/groups/ivc/data.php has a bunch of things, and for medical imaging each field has their favorites.
Ground truth for English word meaning from a number of dimensions is WordNet, but there are also more specialized tools for natural language processing as well.
Neural networks have been reinventing and in many cases replacing and creating new ground truth datasets for various needs constantly.
It is all malleable and changeable and can equally be depended upon or ignored as one wishes, but I want to give a big shout out to reference data, particularly when it’s a herculean effort by many over long periods of time in an attempt to create useful standards to compare with and against, even if it is for the purpose of criticism or replacement.
To show “This is true” in a maximally faceted, deep way, with all the data to back it up, is gutsy and practical.
It may be wrong but it’s useful to have SOME reference to bounce off of just the same. Such is the power of an ontology.