Import/Catalogue/Sabu Raijua building import
Add missing large buildings in Sabu Raijua and Palau Raijua islands, where it is possible to do so with good quality. Approximately 23.000 buildings larger than 25 sq meters and disjoined from 7.000 OSM ones will be imported.
Third version of global building south Asia coverage was published in May 23. Buildings generalization has been applied by source on aerial images not older than 2022-05-15 (date from Google Earth as suggested in FAQ). Currently planet.osm features some 7k buildings in the area. Some Sabu Raijua AOI statistics:
- Buildings (total in AOI]: 52574
- Buildings with >0.752 confidence: 34311
- Buildings larger than 25 square meters: 33100
- (2.) AND (3.): 23641 proposed for import
Data source site: http://sites.research.google/open-buildings/
Data license: https://opendatacommons.org/licenses/odbl/1-0/
Type of license: ODbL v.1
OSM attribution (if required): TBD
ODbL Compliance verified: yes
This is a manual import: sample verifications and corrections will be performed using JOSM. The dataset will be imported through a series of changesets.
The data will be manually reviewed on sample areas, particularly dense ones.
Data Reduction & Simplification
In QGIS the following actions will be performed:
- small buildings under 25 square meters will be removed,
- buildings under 0.752 confidence index will be removed (<90% precision as suggested in FAQ)
- redundant tags (plus_code, confidence) will be removed,
- buildings intersecting OSM ones will be removed.
- self intersecting building will be joined (Shift+J)
- building-highway overlappings (78) will be evaluated (Validator)
Work files and OSM upload candidates will be stored in Github repository.
Latest upload candidate, featuring confidence>0.752, area>25, self-crossing validated, highway-building crossings validated.
The source data is presented as a WKT. This Well-Known-Text consists of a collection of point coordinates that represent each building.
The source tags are:
- confidence, used to remove unreliable best AI guesses (less than 0.752)
- plus_code, not used (see discussion)
- source, not used (set as import changeset tag).
The tag(s) will be as follows:
- building will be the default tag (building=yes).
Changesets will be tagged with:
- source=Google Open Buildings
Data Merge Workflow
Conflation will be resolved using QGIS by automatically removing intersecting OSM buildings. Hence all existing buildings will be preserved.
The import will be a gradual process done by:
If you would like to contribute, send me a message.
Step by step instructions:
- Download the data file 2c5_buildings.csv.gz from Google Open Buildings
- Load data in QGIS as CSV WKT
- remove data outside AOI using [-10.6511607,121.5074158,-10.4027283,122.0127869] bounding box
- remove candidated with area less than 25 square meters
- remove candidates intersecting OSM buildings
- optionally remove buildings with confidence index lower than 0.6 (could be done in JOSM later)
- Load data in JOSM
- inspect on sample areas, preferably dense ones
- Load aerial imagery and manually resolve circular or complex shape buildings (usually large buildings)
- tag object as building=yes then run JOSM validator and solve related issues
- load OSM buildings with AOI query
- resolve validator issues (mostly warning about building-highway crossings)
- Upload the changeset in OSM
In case of any problem with the import, the changeset will be reverted using the JOSM reverter plugin.
See #Data Merge Workflow.
The message to the OSM community general talk was sent on 2023-07-14 and can be found in the forum archives: . It is intended as subject of discussion.
Imported on 2023-07-26, see OSMcha filter.