Global Built-up Surface Validation Grids from Crowdsourced Campaign

50,000 random stratified sample locations were collected via a Geo-Wiki crowdsourcing campaign for validating global built-up surface products. Each location is an 80x80m grid subdivided into 64 cells, with labels for built-up, non-built-up, or unsure status derived from high-resolution satellite imagery. The dataset was created by Linda See and includes raw data, quality control fields, and processed files.

Use Cases

• Validate the accuracy of global built-up surface layers by comparing model outputs to the majority crowd labels in the quality-controlled centroid files.
• Analyze label uncertainty and consensus using the raw cell-level data (Geo-WikiBuilt-upCellsAll.csv) for all 64 cells per grid.
• Assess data quality and contributor reliability using the provided control point fields (e.g., Geo-WikiBuilt-upCentroidsControls.csv) that match crowd answers against expert interpretations.
• Design stratified sampling for validation studies by utilizing the mapping between grid locations and sampling strata provided in Strata.csv.
• Train machine learning models for built-up area detection using the 80x80m grid labels and associated centroid coordinates as ground truth.

Strengths

• Contains a large, structured sample of 50,000 locations designed for global validation.
• Provides multiple data views including raw cell-level labels (64 cells per grid) and aggregated centroid summaries for flexibility.
• Includes explicit quality control mechanisms through control points and fields indicating match with majority crowd answers.
• Uses a random stratified sampling design to ensure representative coverage across different strata.

Limitations

• The sample size of 50,000 locations may be insufficient for highly granular, region-specific validation tasks.
• Label reliability depends on crowd contributor consistency and the resolution of the satellite imagery used.
• Temporal coverage and the recency of the satellite imagery base are not specified, which could affect relevance for current validation.

Provenance

Source

Data collected via the Geo-Wiki platform through a dedicated crowdsourcing campaign.

Collection Method

Visual interpretation of very high-resolution satellite images by crowd contributors, with quality control via expert-provided control points.

Geography

Global