The tradeoff is that this is very expensive at the scale of large geospatial data models both in terms of performance and storage. In practice, it is much more

The tradeoff is that this is very expensive at the scale of large geospatial data models both in terms of performance and storage. In practice, it is much more common to just take regular snapshots of the database. If you want to go back in time, you have to spin-up an old snapshot of the database model. A less obvious issue is that to make this work well, you need to do time interval intersection searches/joins at scale. There is a dearth of scalable data structures and algorithms for this in databases.