Sometimes I feel like I live on Google Earth. Here at concept3D I have overseen and managed the creation of tens of thousands Google Earth models. The models range from a single house in the countryside to 1000s of models within close proximity with one another such as a cluster of city models or a theme park. Part of the creation process is aligning models to the imagery and terrain on Google Earth. Realigning one model to updated imagery and terrain may take up to 30 minutes. To keep projects updated to the most current imagery and terrain becomes a challenge not only to us, but the entire industry using Google Earth. Currently the only solution for realigning a model to updated imagery is to import new imagery and manually move the model to align with the new imagery.
Terrain data is updated independent from imagery updates. This update is the reason that models appear to be floating or sinking into the terrain. This not only affects user generated content, but also Google content as well. If you look at the bottoms of buildings in cities you may find floating buildings. This is usually means that the terrain data was updated after the creation of the models.
I would suspect that Google could automate a process that moves buildings down under the terrain, preventing them from floating, but this does not change the fact that main entrances may also be moved under the terrain. At the moment it is a manual process to update every model in Google Earth to the new terrain and imagery. It is only a manual process because it is hard to have automated a process that moves doors, trees, and other elements up and down to the terrain. Since imagery updates are taken at different angles, this also effects building foot print locations.
In Google Earth, imagery, terrain and 3D modeling are closely connected; better communication with the public on upcoming changes, along new processes will allow efficiency in submitting user generated content.