What we did
Two independently orbiting passes over the same reference site can disagree in a way that looks like a step change. This research area asks which of several candidate causes explains a step like that, and whether the same cause applies everywhere it is seen.
At four reference sites, real per-swath viewing geometry (extracted from satellite product metadata) and real per-band detector identity (extracted from the satellite's own detector-footprint data) were compared against where a step in the reading does and does not appear. At a fifth site, the ground-reference target itself was examined directly against the satellite's own imagery.
What we saw
One candidate cause, how the atmospheric processing handles viewing angle, was ruled out everywhere it was tested: the effect it predicts is tens of times too small and points the wrong way.
A second candidate, a fixed offset tied to a specific hardware detector, was tested by checking whether sites sharing the same physical detector pairing also share the step. At one site, the shared detector pairing did NOT share the step seen at another, ruling out a fixed detector offset as the explanation there and leaving surface-dependent effects as the surviving explanation. At a second, stepping site, no other site in the data set shares its detector pairing, so this particular test could not be run there at all; that site's cause remains open.
At a sixth site, the offset had a different, fully identified cause: the published reference measurement describes a small, engineered target, while RefCal's reading averages over a much larger area of surrounding natural desert. Once that mismatch is accounted for, the site's contribution to the wider picture changes: it goes from looking like the single worst-performing site to a site whose remaining gap is explained by exactly this known mismatch.
Across all sites checked, the offsets found are consistent over time (repeat measurements agree with each other), but they differ from site to site in both size and likely cause. No pattern found so far applies uniformly across every site.
What it means
RefCal can now rule specific causes in or out at specific sites using real data, rather than treating every disagreement as one undifferentiated "the numbers are off" problem. That is real progress: it is the difference between not knowing why a reading is off and knowing, site by site, which explanations are ruled out. It also means no single treatment that applies everywhere is justified yet. Some sites still have open questions, and settling them is the next step in this research area.