Reference
Validation metrics¶
SpectralBridge writes QA metrics so reflectance, correction, harmonization, and masking decisions remain reviewable after a run instead of being hidden behind a single success flag.
Visual QA
PNG and optional PDF outputs summarize the same behavior in human-readable form.
Machine-readable QA
JSON sidecars preserve the numeric summaries for dashboards, regression tests, and automated checks.
Restart-safe review
Validation is part of the workflow contract, not just post-hoc reporting.
Reflectance
What the core spectral metrics capture¶
- bandwise minimum, maximum, and median reflectance
- negative reflectance fractions
- high-reflectance fractions above the QA warning thresholds
- distribution summaries used in the QA PNG and PDF panels
These metrics help detect correction instability, unexpected scaling, or masking problems before the data moves into downstream ecological analysis.
Masks and coverage
How valid-pixel availability is tracked¶
Mask coverage
Mask and invalid-pixel summaries show how much of a scene remains usable after filtering clouds, shadows, water, snow, or no-data regions.
Sampling support
Drone QA also records sampled-support diagnostics so spectral summaries can be interpreted in light of how many valid pixels actually contributed.
Correction metrics
How pre and post correction are compared¶
Correction QA focuses on how much the BRDF and topographic stages changed the data and whether those changes look plausible.
- signed reflectance delta summaries
- median absolute change summaries
- distribution spreads by wavelength
- spatial summaries of where corrections were strongest
- flags that distinguish no-op behavior from meaningful change
Large changes are not automatically wrong, but they should be interpretable in the context of scene geometry, terrain, and QA overlays.
Harmonization metrics
How translated sensor products are checked¶
Brightness adjustments
When brightness normalization is active, the applied coefficients are recorded so downstream users can see that the translated product was not just a raw convolution.
Spectral alignment
Convolution and harmonization diagnostics track how corrected reflectance relates to the target sensor bandpass assumptions and whether the resulting products look internally consistent.
Geometry and context
What helps explain unusual scenes¶
Geometry and ancillary summaries make the QA outputs interpretable instead of leaving users with unexplained deltas.
- solar zenith and azimuth context
- view geometry summaries
- slope and aspect context where relevant
- drone correction status and overlap diagnostics
Using the metrics
When a flight line deserves closer review¶
Common red flags
Large brightness shifts, high invalid fractions, unstable correction summaries, or severe out-of-range reflectance all deserve a second look.
Best follow-up step
Open the QA PNG or PDF and compare it with the JSON sidecar rather than changing coefficients or masks blindly.
Where to go next