Ground control points (GCPs) are precisely surveyed marks, visible in imagery, that anchor a photogrammetric or LiDAR project to real-world coordinates. Checkpoints are their independent siblings — surveyed marks withheld from processing and used only to test the result. Together they are the accuracy machinery of mapping: control creates it, checkpoints prove it, and the RMSE report communicates it.
Control versus check: the crucial split
GCPs participate in the adjustment — the block is bent to fit them, so residuals at GCPs flatter the truth. Checkpoints stay outside the computation; the differences between their surveyed and mapped positions measure genuine achieved accuracy.
Standards such as the ASPRS positional accuracy framework formalise this: accuracy classes are claimed from checkpoint statistics, with minimum checkpoint counts and distribution rules.
How many, and where
Numbers depend on block geometry and positioning method: an uncontrolled consumer-drone block may need a dozen well-spread GCPs; an RTK/PPK block needs far fewer for control but still wants checkpoints everywhere accuracy will be claimed. Corners-plus-centre is the minimal pattern; corridors need control leapfrogging along their length; elevation-critical work adds vertical control in flat, featureless areas.
Placement quality beats quantity: marks must be unambiguous in imagery (high-contrast targets), stable, and surveyed to accuracy several times better than the product specification.
Reading an accuracy report
A serious deliverable states: datum, projection and geoid; control and checkpoint counts and layout; RMSE horizontal and vertical at checkpoints; and the standard/class claimed. Beware reports quoting only software-internal reprojection errors — those measure self-consistency, not truth.
For buyers, the one-line test is simple: “show me checkpoint residuals against independent survey.” Vendors doing real measurement answer instantly.