Combining sensors
Objective
Combine sensors so that a useful conclusion about the ground becomes more confident, for example that a low value is unlikely at a given wavelength, even when several higher values remain possible.
Observed in real data Current focus See the results How stages are defined
Extending the two-sensor comparison to more locations and dates, on an equal footing.
Sensors differ in spectral response, spatial resolution, overpass time and calibration history, and their errors are not independent: shared references and shared processing can make two sources look like more evidence than they are. Counting the same evidence twice produces confidence the data do not support.
This is usually the real question behind satellite data: taken together, do the sources support the signal or not? The answer is often not one exact value, but a clear statement of what is and is not likely.
At the same location and wavelength, we compare the confidence each sensor gives on its own with the combined result, including cases where sensors duplicate each other, share errors or are degraded, and we check against independent reference measurements where they exist.
Combined confidence rises only as far as the independent evidence justifies, with shared errors carried through into the result.
Each result is dated and published on its own page with its data, conditions and limits. Results that went against the objective are listed here with the rest.
-
Putting Sentinel-2A and Landsat 9 into the same band space
RefCal can place official Sentinel-2A and Landsat 9 response models over the same real measured spectrum and preserve a reproducible band-by-band comparison. This is real-data evidence for the band-space bridge. It is not evidence that the two satellites agree radiometrically on observed scenes.
Full result, data and credits ->Fig. 1 / Landsat 9 minus Sentinel-2A, by band Real data · one RadCalNet spectrum- Sentinel-2A, the reference
- Landsat 9, relative to it
Each value comes from one measured RadCalNet reflectance spectrum from RVUS, forwarded through the official Landsat 9 OLI-2 and Sentinel-2A spectral response models on the same wavelength grid, Landsat 9 minus Sentinel-2A.
No agreement tolerance is published for these differences, so none is drawn. Data: RadCalNet (Bouvet et al. 2019), Copernicus Sentinel-2, USGS Landsat 9.
Established work this area builds on. Results from the programme are published separately, with their data and limits.
- Determining uncertainties FIDUCEO project, University of Reading Why errors shared between measurements must not be counted as independent evidence.
Questions about this work?
If you work on this problem, or hold reference data that bears on it, contact us.