Processing decision guide¶
First decide what you have¶
- Identify the physical quantity. Read product-level and unit metadata, not only the variable name.
- Check geometry and quality. Locate required sidecars and determine whether data are already mapped.
- Choose an input branch. Surface reflectance normally bypasses atmospheric retrieval. Supported at-sensor inputs can enter the optional ISOFIT path.
- Choose surface corrections from evidence. An available function is not a reason to apply it.
- Prepare the analysis grid and spectral support. Align only when needed and document resampling.
- Calculate features and export provenance. Verify the written output, not only an in-memory plot.
Three typical routes¶
| Route | Sequence |
|---|---|
| Provider L2 reflectance → spectral features | Open → inspect → quality policy → optional spatial/spectral preparation → features → export |
| Airborne reflectance group → normalized windows | Open group → sample selected regions → per-line topo diagnostics → optional topo → grouped FlexBRDF → apply → seam/angle diagnostics → export |
| Supported L1 input → mapped reflectance | Open → prepare ISOFIT inputs → retrieve → optional satellite NBAR → map → export cube, retrieval layers, QA, and provenance |
Corrections are not interchangeable¶
Atmospheric correction addresses atmospheric effects on the measurement. Topographic correction addresses illumination associated with terrain. BRDF normalization adjusts directional reflectance to a selected angular condition. Georeferencing addresses spatial placement. Smoothing changes spectral shape. None is a substitute for another.
The implemented high-level topographic/FlexBRDF pipeline is airborne-only. That implementation boundary must not be read as a scientific assertion that terrain does not affect satellite imagery. Similarly, the MCD43 route is the package's supported satellite strategy, not proof that all satellites lack useful angular diversity.
Avoid automatic overcorrection¶
Provider products may already include corrections relevant to your target analysis. Inspect their processing documentation and your residual diagnostics. A topographic verdict of skip, refuse, or inconclusive is meaningful information. Do not convert all of those results into force_topo=True for convenience.
Reproducible decisions¶
Keep input identifiers, processing region, sample strategy, masks, coefficients, solar reference, SRF source, interpolation choices, cache provenance, and dependency versions. If you transfer coefficients from provider reflectance to a new retrieval, label that transfer and validate it separately.