Ground Control

The planning questions in Chapter 8 asked where the ground control points and fixed landmarks are. This chapter answers how their coordinates get measured, why the answer is usually RTK, and whether the map needs them at all.

Precision, accuracy, and what they cost

Start with three words that get confused in the field.

Term	Question it answers
Precision	Do repeated readings of the same point agree?
Relative accuracy	Do the points in the map agree with each other?
Absolute accuracy	Do the points in the map agree with the earth?

A receiver can be precise without being accurate. Take five readings at the same landmark. If all five land within 2 cm of each other, the receiver is precise, even if every reading sits 2 m from the truth. The same split applies to a finished map. The SLAM guarantee from Chapter 11 is internal consistency: points that agree with each other to within centimeters. That is relative accuracy, and a map can have it while still being shifted, rotated, or tilted relative to the earth. Only absolute accuracy makes a map interlock with the world outside it. Whether the application needs that interlock is the decision that drives this chapter.

A phone-grade GNSS receiver reports a position accurate to 1 to 5 m under open sky. That is enough to find the site. It is not enough to anchor a map: 5 m of error at the gate is the difference between the right road and the wrong one.

Method	Typical horizontal accuracy
Standard GNSS	1 to 5 m
Differential GNSS	0.3 to 1 m
RTK, fix solution	1 to 2 cm
Total station survey	millimeters, relative

How RTK works

RTK, Real-Time Kinematic, is differential GNSS pushed to its centimeter limit. Two receivers work together. A base station sits at a known position, on site or at a regional reference network, while a rover moves across the survey area. Both receivers track the same satellites, so they suffer the same errors: atmospheric delay, satellite clock drift, slight wander in the broadcast orbits. The base knows its own position, so it computes the error on each satellite signal in real time and streams a correction to the rover, over radio or over the internet through NTRIP, the standard protocol for streaming RTK corrections. The rover applies the correction, and the shared errors cancel. Corrections work best over short baselines, the distance between base and rover. 10 to 20 km is the practical limit, and accuracy degrades first in the vertical.

On an HD mapping rig, the rover is usually not a separate instrument at all. It is the GNSS receiver built into the scanner vehicle or the handheld unit, so the corrected trajectory and the scan data are captured in the same pass.

The centimeter step is carrier phase. A receiver measures the code on a GNSS signal to a few meters. It can also count cycles of the carrier wave underneath, each a fraction of a meter long, which measures distance far more finely. The problem is ambiguity: the receiver knows the fractional phase of the carrier but not how many whole cycles came before. RTK resolves those integer ambiguities while the rover moves, using the slowly changing satellite geometry. When the solution locks, the receiver reports fix, and the position is good to 1 to 2 cm horizontally and 2 to 4 cm vertically. Before the lock, the receiver reports float: smooth-looking, decimeter-level, and not ground control. A float reading looks like a good reading. It is not one.

RTK also demands open sky. Multipath, the same phenomenon that turns container walls into ghost surfaces for LiDAR, biases GNSS too: reflected signals arrive late, and the rover’s solution shifts. Choose landmarks with a clear view of the sky. At Harbor Yard, the six landmarks were picked partly for this reason: a stop sign at the edge of an open lane, a street light on open pavement, a gate post at the site entrance. A corner deep inside the container canyons is a poor RTK site no matter how prominent it is in the point cloud.

A corridor survey through an urban canyon faces the same problem at scale: tall buildings, overpasses, and dense tree canopy drop or reflect the signals for minutes at a stretch. Plan for the outage rather than hoping for a fix. A high-grade IMU can coast through a short gap on its own, and a wheel-speed sensor, called a DMI for distance measuring instrument, keeps the trajectory honest when satellites disappear entirely. Handheld work indoors or under dense canopy is the extreme case: the IMU runs blind, and only tight ground control spacing holds those sections straight.

Does the map need an anchor?

So what is all this accuracy for? A ground control point anchors a map to the earth. It is a marker whose coordinates are known in a shared frame, so every point measured against it inherits world coordinates. If the map must share coordinates with anything outside itself, GIS layers, engineering plans, a legal boundary, a second dataset captured elsewhere, that anchor is not optional. Sharing a coordinate frame with the world is the case for absolute accuracy, and it is the case the georeferencing step in Chapter 10 assumes.

Not every application makes that demand. If the map only has to agree with itself, ground control is a convenience rather than a requirement. A vehicle that will operate only against this one map, never against anything external, needs the map’s points to agree with each other. Whether the whole cloud sits 2 m from true north changes nothing, because nothing outside the site will ever measure it. The same applies to pure measurement jobs: stockpile volumes, dock clearances, distances between fixed features. Relative accuracy is the requirement, and SLAM plus RTK readings at fixed landmarks delivers it without a formal anchor.

There is a middle option for self-contained sites that still want centimeter consistency: relative RTK. Set the base station over a point whose coordinates you do not know. Let it average its own position for 5 to 15 minutes, and it converges on a stable position that is accurate to a few meters and offset from the earth by the same few meters. Every rover reading inherits that offset. The survey is centimeter-consistent with itself and shifted as a whole, which is exactly what a self-contained site needs, at no surveyor cost.

Ground control at Harbor Yard

Harbor Yard sits at the demanding end of the spectrum. The map had to agree with the earth, so the full ceremony applied. The base station stood over the survey nail for the day. The nail’s coordinates were known from the surveyor’s work, so the base knew its position the moment it switched on. The rover was built into the scanning rig on the scanner vehicle, so there was no separate occupation of the landmarks: as the scanner drove the loop, the rig logged its trajectory in RTK fix, and the cloud came off the vehicle already carrying world coordinates accurate to 1 to 2 cm in UTM zone 10N. The six landmarks were not independently measured. They were picked out of the finished cloud, sharp features the scanner had already captured, and their coordinates read from the georeferenced data itself. The landmarks are located by the scan, not surveyed beside it.

The value of all this is practical. Ground control does not have to mean a separate survey campaign. The survey nail at Harbor Yard, set and surveyed by a licensed surveyor, provides the formal anchor for the wider site, and a total station remains the tool for work that needs millimeter precision. But the centimeter coordinates at the landmarks came from the scanning rig’s own receiver during the normal capture drive. No pole, no second crew, no second site visit: ground control as a byproduct of scanning.

What comes next

The next chapter covers scan processing: filtering the raw cloud, registering multiple scans into one, georeferencing the result to world coordinates, and the quality checks that gate the first traced vector.