Nothing Is Final: Improving a Base Position After the Fact
A base established quickly, with a modest method from earlier in this companion, does not have to be treated as a permanent limitation on everything collected against it. If that coordinate is later refined, a real, practical path exists to carry that improvement backward into work already done.
When a base's original coordinate is later replaced by a more accurate one, the known difference between the two can be applied as a translation to rover positions already recorded during that session, a technique that works because a base coordinate error passes through RTK's differencing untouched rather than being absorbed by it.
Why This Works, Precisely
Recall the mechanism from earlier in this companion: a rover solves its position by differencing its own observations against the base's, canceling shared atmospheric, orbital, and clock error. That differencing process never actually checks or absorbs the base's assumed coordinate. It simply treats that coordinate as the fixed point everything else gets attached to. A change in that one number, discovered later, therefore propagates through to every rover position from that session in very nearly the same way, uniformly, since nothing about the differencing itself depended on that coordinate being correct in the first place.
The correction vector used here is not something the RTK engine computed and stored during the original session. It is built afterward, simply the difference between the base's original assumed coordinate and its newly refined one, then applied as a translation to whatever positions were recorded under the old assumption.
Where This Approximation Holds, and Where It Strains
Applying a single, uniform shift to an entire session's worth of rover positions is an excellent approximation for a typical short baseline and a modest coordinate correction, the situation this technique is actually built for. It is not a mathematical guarantee valid at any distance or any size of correction. A very long baseline, or a substantial change in the base's coordinate, introduces enough non-uniformity that a simple shift stops being a safe assumption, and the technique should be treated as a practical tool for the ordinary case, not an unconditional rule.
This correction technique does not work because RTK secretly computed and saved a baseline vector during the original session. It works because a base coordinate error passes uniformly through the differencing process at typical baseline lengths, which is a different, and more limited, guarantee than treating the correction as a stored, exact quantity.
The Alternative, When It Exists: Full PPK Reprocessing
If the rover was also logging its own raw observations throughout the session, rather than only keeping its computed, real-time positions, a more rigorous option exists: reprocess the entire session from scratch using the base's newly refined coordinate, properly re-deriving the relationship between base and rover rather than approximating it with a single translation. This costs considerably more effort than applying a shift, and it is the more defensible choice whenever the baseline was long, the correction was substantial, or the resulting coordinates need to withstand serious outside scrutiny.
A base's coordinate is refined a week after a short, close-range survey, and the correction is small. Is a full PPK reprocessing of the rover's raw logs necessary here? Probably not. A simple translation, using the difference between the old and new base coordinates, is a reasonable, well-supported approximation at this scale, and full reprocessing would be effort spent well beyond what the situation calls for.
A base coordinate discovered to be imperfect after the fact does not require redoing the work collected against it. The difference between the old and new coordinates can be applied as a translation to that session's rover positions, a reliable approximation at typical baseline lengths and correction sizes. Full PPK reprocessing, available only if the rover logged raw data, is the more rigorous alternative when the baseline was long or the correction was substantial.
Everything in this companion has assumed building a base is the right answer in the first place. The next chapter asks when it genuinely is not.