Quality flags and masking

The QA layer translates available provider masks and selected derived conditions into a uint16 bit field. A pixel can carry several flags simultaneously. Original provider layers remain useful for interpreting how a flag was obtained.

q = hp.quality_flags(ds, derive=("fill", "terrain_shadow", "steep_terrain"))
print(hp.quality_table(q))
masked = hp.quality_apply(ds, q, drop=("fill", "cloud", "cloud_shadow", "cirrus"))
shadow = hp.quality_decode(q, "terrain_shadow", "cloud_shadow")

Bit definitions

Bit	Flag	Meaning
0	fill	No observation / nodata
1	saturated	Saturation condition available from the product
2	cloud	Opaque cloud
3	cloud_shadow	Cloud-shadow flag
4	cirrus	Cirrus flag
5	snow_ice	Snow or ice
6	water	Water flag
7	haze	Haze/aerosol-related provider flag
8	sun_glint	Glint condition
9	terrain_shadow	Nonpositive local solar-incidence cosine
10	steep_terrain	Slope exceeds the selected threshold
11	ac_failed	Atmospheric retrieval failure indication
12	brdf_filled	BRDF handling used missing/substituted support
13	negative_reflectance	Selected fraction of usable bands below zero

Build is not mask

quality_flags() reports conditions. quality_apply() masks the selected main variable where specified flags occur. Building steep_terrain does not exclude those pixels unless that flag is in drop. Choose a policy appropriate to the analysis rather than assuming all bits are equally disqualifying.

Default derived flags are fill and terrain_shadow. Optional spectral cloud screening is not automatically run in every reader or workflow. A zero-valued cloud bit can mean that no cloud information was available; it is not a cloud-free certification.

Export categorical values correctly

Use nearest-neighbor or another categorical-safe method for flags, never ordinary averaging. Include the bit definitions in output metadata. Review whether geometry-derived flags remain valid after resampling and whether source flags were present in the first place.

See quality API and export.