The idea of a GML observation (see other notes in this blog) was conceived as a model of the "act of observing" – the doing of it. While it has been fairly widely applied for sensor data, and even for tourist photographs, the concept equally applies to the usual update of geographic features based on land surveys, GPS or photogrammetry.
Consider the example of a crew of a water supply company out surveying a water main. They are responsible for the accurate location of the water main which wil be recorded in the company's geographic database. In the course of doing this survey they "observe" that one of the parcel boundaries is in error or at least with respect to their information baseline. They record this for company records. They would like to update the City's database, but that is not their responsibility and in most jurisdictions there is no formal mechanism to make the update. Even if there were, what should they report? They cannot change the location of the parcel for the city. Even if they had the authority, their "observation" may need to be modified (or the existing parcel fabric may need to be modified) before their observation can be integrated into the City's geographic database. They can, however, submit a GML observation recording the time, the observed feature characteristics (i.e. the parcel boundary geometry), and the means of making the observation. No one can argue with this information and it is quite conceivable that this information could be directed to and recorded by the city. The city could then process this GML Observation (provided as a WFS transaction) and then use it to create a transaction that modifies the actual parcel fabric.
Observations can be useful items.