Evidence date7 Aug 2026
AreasP06
OutcomeDid not hold
●   How we do it / Result

First claim-grade confidence check did not meet its target

Question

Do RefCal's published per-band confidence intervals contain independent reference values as often as intended across a claim-grade real-data set?

Did not hold Superseded How stages are defined

This result has been superseded by Confidence coverage on a like-for-like surface basis (22 Sept 2026). It stays published as part of the record.

§ 01
The result

What we did

RefCal compared its published per-band confidence intervals with values then treated as RadCalNet surface reference measurements. The claim-grade subset contained 131 measurement rows across 121 independent site-days at all six available sites.

What we saw

On that basis, the intervals contained the reference value in 0.397 of cases, against a nominal 0.683. Later lineage work established that the comparison mixed surface reflectance with top-of-atmosphere reference values, so this number does not describe current surface-to-surface confidence performance.

What it means

This run failed on the basis used at the time and remains part of the dated record. It taught a more fundamental lesson than the original verdict: confidence coverage is not interpretable when the product and reference describe different measurement levels. The corrected surface-to-surface result now carries the current evidence.

§ 02
Conditions and limits
  • This is the historical verdict produced when SIAC surface reflectance was compared with a RadCalNet value later found to be top-of-atmosphere reflectance. It is not the current confidence verdict.
  • The coverage factor was declared as one because RadCalNet does not publish it; the register therefore keeps the overall verdict qualified even though the measured shortfall is large.
  • This is an outcome-level result. It states that confidence is provided per band but does not describe how RefCal produces or updates those intervals.
  • The result remains visible because it was the first claim-grade run and shows why matching the reference basis matters. A corrected surface-to-surface result supersedes it.
§ 03
Data and credits
  • RadCalNet RadCalNet reference measurements, with citation to Bouvet et al. 2019, Remote Sensing 11(20), 2401.
  • Copernicus Sentinel-2 Contains modified Copernicus Sentinel data 2024-2025.
§ Contact

Questions about this result?

If you work on this problem, or hold reference data that bears on it, contact us.

Your message is emailed to info@spectraworks.nl. See the privacy policy.