Missing buildings in a few areas

[chaitanyach]

I was comparing the July 2024 buildings, and I found some square tiles missing most of the buildings when compared to May. These tiles are at zoom level 15 of WMTS with Pseudo Mercator projection.
Here are some examples:

Most of the buildings missing in the tile at 26.268458N,80.107122W. They were present in May release.

A few tiles in a row near 32.80196N,117.17697W are missing most of the buildings (quadkey 023013221210301). Present in May release.

There are more such areas.

[skmoore]

@jenningsanderson I see this too, one example that was present in May and missing in June and July is 08b44a12b155efff020051dffb3f954d

[jenningsanderson]

Interesting. I don't see these features in OSM nor in Microsoft ML Building footprints. It looks like 08b44a12b155efff020051dffb3f954d was an MSFT building from May but is not present in the newer data available from Microsoft.

Here's what I see in Rapid with MSFT ML turned on:

Odd that this particular square is empty. I can't seem to find deleted OSM features that used to be there, either.

[chaitanyach]

The missing buildings are missing in the 2024 August release as well and in the same areas. I checked the two examples I mentioned in the beginning.

[jfmartinez4]

I commented on #216 about missing data, and I wasn't sure if it was the data or something on my end.

[jenningsanderson]

I commented on #216 about missing data, and I wasn't sure if it was the data or something on my end.

Here's the link to the explore site where the data is visible: https://explore.overturemaps.org/#14.11/38.091/-122.23557

Perhaps try using DuckDB or the Python CLI utility to download the data for this region again?

[jenningsanderson]

This particular issue seems to be coming from the MSFT ML buildings dataset — buildings that were present before are no longer there.

[chaitanyach]

@jenningsanderson, if a source dataset removes some buildings, are they assumed to no longer exist even if they exist in other sources?

[jenningsanderson]

@chaitanyach It would depend on which source removed them. In this case, we don't have another source that has the buildings, so they are gone. However, current conflation logic would have it that if a building was removed from a higher priority source (such as OSM / Esri), but still existed in a lower priority source (Microsoft, Google), then we would choose that building because we currently prioritize coverage. We recognize the error that this could introduce and are working on considering a building being removed from a higher priority source as a signal to not re-add the building from a lower priority source. Hope that helps!