We need a way to ensure that should a dataset be converted multiple times (eg. due to an update to the conversion process), the identification generated for each building is identical every time – this should be the case even if different people ran a conversion on the same dataset at different times in different places on different machines.
At a more advanced level, we should find a way to try and keep this identification even when the dataset has changed slightly (eg. an update from the data author), or if we use another dataset from a different author that describes the same building.
For example, a city may release a complete upgrade to their CityGML data that dramatically changes building objects and uses different IDs to the previous version – we should try to make a best guess on whether we already know about the updated building, perhaps by comparing location data and the overlap of each footprint. If enough things correlate then it's likely the same – the approach doesn't need to be 100% sure.
Building attributes that may be useful for identification
- Location / coordinates
- Address
- Build date
- Area
- Volume
- Who's on First neighbourhood / area ID
We need a way to ensure that should a dataset be converted multiple times (eg. due to an update to the conversion process), the identification generated for each building is identical every time – this should be the case even if different people ran a conversion on the same dataset at different times in different places on different machines.
At a more advanced level, we should find a way to try and keep this identification even when the dataset has changed slightly (eg. an update from the data author), or if we use another dataset from a different author that describes the same building.
For example, a city may release a complete upgrade to their CityGML data that dramatically changes building objects and uses different IDs to the previous version – we should try to make a best guess on whether we already know about the updated building, perhaps by comparing location data and the overlap of each footprint. If enough things correlate then it's likely the same – the approach doesn't need to be 100% sure.
Building attributes that may be useful for identification