GCP & PPK Accuracy

PPK gets your photo positions right; ground control points (GCPs) anchor the whole reconstruction to surveyed coordinates; checkpoints prove the result. This guide covers all three. (For the upload flow itself, see Running a Survey.)

Getting control points in

Open your survey run and hit GCP Editor (it shows a count badge once points exist). Three ways to load control:

• Import CSV — import surveyed control from a DJI / D-RTK 3 export. The importer recognizes id/name columns, Northing/Easting (or X/Y) or Lat/Lon, and Z/Elevation/Height; you set the Source CRS (EPSG code) and whether heights are orthometric (GEOID18/NAVD88) or ellipsoidal (GPS/WGS84).
• Recover from RTK — scans your uploaded D-RTK 3 .mrk rover captures and recovers the FIXED marks as control points.
• gcp_list.txt — a classic GCP text file (EPSG header + points) dropped into the upload zone with the photos is auto-detected.

You can also add a point by hand in the editor.

Marking GCPs in the photos

A GCP only helps once the software knows where it sits in your images. In the GCP Editor: pick a control point on the left, choose a photo from the strip, and click the target in the image. A few marks across different photos per point is the goal.

Find My GCPs does the tedious part: AI finds each imported GCP in your photos and proposes the pixel marks — you confirm every one (there’s an “Accept best detections” shortcut for the confident ones). Nothing enters your GCP file without your confirmation.

There’s also Auto-Detect, a blind computer-vision scan of all photos for ArUco markers and checkerboard patterns — useful when you flew coded targets.

Hit Save and the control is attached to the run; process (or reprocess) and the reconstruction is anchored to it.

Proving accuracy with checkpoints

Checkpoints are surveyed points held out of processing and measured against the finished surfaces — the honest, independent test.

On the map workspace (Open Map), the Checkpoint Accuracy section can Pull NGS Monuments — published survey control near your site — and add selected monuments as check points. Then Run Accuracy Check (it needs the elevation model) measures each checkpoint against your results. A vertical shift field lets you reconcile a datum offset (ellipsoidal vs orthometric) if your checkpoint heights use a different reference.

The result reports, in feet:

• Horizontal RMSE and Vertical RMSE
• NSSDA accuracy at 95% confidence for each
• Per-checkpoint residuals, plus a plain-language statement

The same accuracy section appears in the run’s Deliverable Report — checkpoint counts, RMSE values, NSSDA 95% figures, vertical bias, and the per-checkpoint residual table — so your client’s PDF carries the proof.

Common questions

GCPs or PPK — do I need both? PPK corrects where the camera was; GCPs pin the model to surveyed ground truth. PPK alone gets you excellent relative accuracy; add GCPs when the deliverable must tie to a survey datum, and checkpoints when you have to prove it.

How many GCPs should I place? Five well-spread points is a common floor for a typical site — corners plus center — with more for large or elevation-varied sites. Keep some surveyed points aside as checkpoints instead of using everything as control.

Why do my checkpoint heights look uniformly off? Usually a vertical-datum mismatch — ellipsoidal vs orthometric heights. Set the heights type correctly on CSV import, or use the vertical-shift field on the accuracy check to reconcile a known offset.

Where do checkpoints come from? Pull NGS monuments from the map’s Checkpoint Accuracy section — published control near your site with known coordinates. Points you import as GCPs are used as control, not checkpoints.