A cloud mask is a per-pixel layer that records which observations in an image have been assessed as cloud, so that users can select or exclude them.12

The CEOS Analysis Ready Data specification for surface reflectance requires metadata that indicates whether a pixel is assessed as being cloud.1 It also asks for a scene-level cloud cover extent, which is a proportion rather than a pixel flag, so the two should not be confused.1

In the Landsat Level 2 quality band, cloud confidence flags come from the CFMask algorithm, and high-confidence cloud pixels are dilated.2 Fmask is a separate, openly documented algorithm that detects clouds in Landsat and Sentinel-2 imagery and recommends a cloud dilation setting per sensor.3

A cloud mask is therefore a product of a chosen algorithm and version, not an observed quantity. The retained sources do not compare the accuracy of the algorithms, so none is claimed.

Footnotes

  1. Committee on Earth Observation Satellites, “CEOS-ARD Product Family Specification: Surface Reflectance”. https://ceos.org/ard/files/PFS/SR/v5.0.2/CEOS-ARD-PFS-SR-v5.0.2.pdf ↩ ↩2 ↩3

  2. US Geological Survey, “Landsat 8-9 Collection 2 Level 2 Science Product Guide”. https://d9-wret.s3.us-west-2.amazonaws.com/assets/palladium/production/s3fs-public/media/files/LSDS-1619_Landsat8-9-Collection2-Level2-Science-Product-Guide-v6.pdf ↩ ↩2

  3. Geographic Environmental Remote Sensing Laboratory, “Fmask”. https://raw.githubusercontent.com/GERSL/Fmask/master/README.md ↩