Monthly on-site flights to monitor stockpiles or haul road progress. Six months in, somebody sees that the numbers look like they all uniformly shifted a little in the same direction across the whole website versus, say, whatever came before it. They didn't move any material overnight, no equipment crashed, and the crew followed their flight plan on autopilot — like they always do. This was nothing fundamentally different on the ground. Neither survey but the invisible reference frame underneath both sites.
An Actual Case — The Correction Source That Spontaneously Developed Without Telling Anyone
Just this was documented by Propeller Aero with their DirtMate RTK hardware. A few VRS (virtual reference station) correction providers allow an "AUTO" mountpoint select which reference station or correction model to use — and that selection can vary between sessions, with no notification to the user. This led to an unexpected role change for two identical flights from a particular job site, despite the fact that each tracked session reported an accurate and precise fix.
Just as a side note, the framing that Propeller uses is worth keeping in mind: high accuracy does not equal correct reference frame. Even if a device can report 0.05 feet of precision, it may be several feet away from the correct absolute position if the underlying correction source changed. Changing the height of poles or other field parameters doesn't solve this problem, because it is not a measurement error within a session — it is a mismatch between sessions.
How Can the Survey be Right but Wrong
Mistaking precision for correctness is a common occurrence as they are two different properties. Precision means a survey is internally consistent and repeatable where correctness means that session has actually been anchored to the same absolute coordinate system as every other survey it is being compared against.
This is directly confirmed by academic testing of RTK positioning: separate sessions cannot be reliably compared at centimetric level solely on the basis of each session's internal fix, since each independently reinitialises. Achieving true centimeter-level agreement between two surveys months apart requires the existence of a physical reference point with known surveyed coordinates where both sessions can be tied back to the same coordinate system in real-world space — no matter what corrections were applied by the network in-between flights.
The Reference Frame Itself Is A Moving Target
This isn't a one-off glitch to look out for — it is part and parcel of the basic theory around satellite positioning actually working. WGS84 is not a monolithic standard, but has been updated through multiple realizations over the last decade due to continual improvement of the underlying station network and modeling, with recent versions including G2139 and G2296. Every new realization is slightly more accurate than the last, but it also implies that the coordinate system itself shifts just a little with every update to a realization and reference frames generally carry an associated year for exactly this purpose. When a receiver, correction network, or processing pipeline adopts the newer realization between Survey #1 and Survey #2, "same" just got real — and there's nothing to tell you that this has happened.
Elevation has its own flavor of this confusion. After inexperienced operators had confused the two, DJI literally coded an update that clearly delineated ASL (using the EGM96 geoid) vs. ellipsoidal height for its Matrice 300 RTK as separate parameters in a firmware update. If an update to an elevation definition on a platform changes which one it defaults to using, then at the change boundary along that timeline falls two different baseline definitions but still have full features of a time series.
How do you actually catch this before it breaks a time series?
More accurate hardware—not in this case; every session in the above scenario was already reporting a clean, high-precision fix. And the fix is to treat repeat monitoring in a different way from a one-off survey. How for work, which relies on comparisons across time periods (i.e. two or more sessions), should make sure no reliance is made on an "AUTO" correction source; Instead, wherever it can be anchored every flight to a persistent, physically monumented benchmark that can get independently re-survey each visit instead of trusting by default that two RTK sessions months apart landed in the same reference fframe.
In addition, it is worthwhile documenting the specific realization of the reference frame and correction source utilized for each survey date, rather than simply retaining the datum record on file. If a site-wide change occurs which appears to be a uniform drifting effect throughout, and isn't localized where it reasonably could've had a real-world basis — then the fact that it has happened is strong in itself point to reference-frame artifact rather than true ground movement.
Implications on the Work Items Relying On It
This remains directly linked to work that is reliant on counting a number of things measured before- stockpile volume tracking labour monthly, or any repeated topographic survey simply used to measure the progress/ground movement fringe from time to time. Just as a single delivery landing in an entirely wrong datum falls within the same overarching problem category, here it can disguise itself within what looks like perfectly normal high-precision survey on both ends. If someone finds that monthly volumetrics jumped in lockstep then yeah that's worthy of a reference-frame check right there—I'd be concerned about stuff going missing, ground subsiding or the crew flubbed something!
Where This Leaves You
The geometry could be perfect on both survey dates, and the difference between them could still be wrong because the invisible reference frame beneath both surveys could never actually hold steady. But for repeat monitoring work — any time series of work built on comparing the today's GIS layer to last quarter's — a physical benchmark (and not one that cares if we updated our software) is in fact what holds the time series together.
