Do we have any list of dates on which these inferencing or footprint overlaps : Inconsistent Detection and Spatial Shifts Due to Imagery Date Mismatches

We’ve observed a pattern of false positives and spatial misalignments when using the Global ML Building Footprints dataset for inferencing and overlay tasks. After investigation, it appears that these issues are often caused by mismatches between the capture dates of input imagery and the reference dates used during model training or footprint generation.

Problem:

The Global ML Building Footprints dataset does not provide metadata on the acquisition dates of the source imagery used for footprint extraction.

When these footprints are overlaid on imagery from a different time period (e.g., more recent or older satellite scenes), it results in:

False detections where buildings are marked but do not exist in the current scene.

Spatial shifts or alignment issues due to landscape or urban development changes over time.

This is particularly problematic when using the dataset in regions with rapid urban growth or significant seasonal variation.

Request:

Could a list of capture dates or imagery metadata (e.g., acquisition date, sensor/platform used) be published or linked for each tile or region covered by the footprints?

Alternatively, could guidance be provided on the best practices for temporal alignment when using these footprints for downstream analysis?

Start by investigating whether acquisition dates or sensor/platform metadata exist for the Global ML Building Footprints tiles or regions. Done would mean publishing or linking that imagery metadata, or documenting practical guidance for aligning footprint data with imagery from different dates.