How the data-quality score is calculated
A 0–100 score from geometry, provenance, flags, duplicates, verification and licence clarity.
Each dataset gets a score out of 100. It is a health check on the data as stored, not a judgement of the source's science.
| Component | Weight | Full marks when |
|---|---|---|
| Geometry present | 30 | No feature lacks a geometry |
| Provenance | 25 | Every feature links to its source record |
| Quality flags | 15 | No feature carries a QA flag (such as a rejected value) |
| Duplicates | 15 | No repeated source ids |
| Verification | 10 | Units and the archived copy have been checked |
| Licence clarity | 5 | The publisher states a licence |
Reading it. Most datasets score in the 90s. A dataset can score 95 and still be unsuitable for your purpose; and one scoring 70 may be perfectly usable if the missing marks are only "licence to be confirmed".
Where to see it. The Data Quality page lists all datasets with their scores, and every dataset page shows its own.
Related reading
More in Methodologies
- Confidence labels: calculated, illustrative, assumed
- How recommendations and briefs are generated
- The official yearly series (1985–2025)
- Repairing invalid geometry
- How features are assigned to countries
- Spatial statistics: Moran's I and Getis-Ord Gi*
- Geodesic area and perimeter
- How Global Mangrove Watch maps are made
Still have a question?
The AI analyst answers from these articles and from the data.
Ask Atlas