Quality

Cloud cover

Cloud percentage matters per-AOI, not per-scene. Thin cirrus is the silent killer.

GOOD · 20% SCENE · 0% AOIAOIusableBAD · 20% SCENE · 100% AOIAOIuselessBoth scenes report 20% cloud cover. Only the left one is usable.
Fig. 1 Cloud cover matters per-AOI, not per-scene. The same scene-level cloud percentage can be acceptable or catastrophic depending on where the clouds sit relative to your area of interest. Always specify cloud thresholds against the AOI footprint, and use ML-based masking to catch thin cirrus.

Cloud cover is usually specified as a maximum acceptable threshold — 'less than 10% cloud.' The trap is that scene-level cloud percentage is misleading: 20% cloud over exactly the wrong 20% of the AOI is 100% useless. Good providers report cloud cover over the AOI specifically, not over the whole scene. Geopera's QA does this.

Cirrus is the silent killer

Thin cirrus often is not caught by rule-based cloud detectors — it looks fine to a threshold but destroys reflectance values for any quantitative analysis. Machine learning cloud masking trained on large datasets catches cirrus reliably; rule-based systems do not. This is why we use ML cloud masking as the default.