RTK photogrammetry without GCPs: when network corrections are enough

Ground control points are the default answer to "how do I trust a drone map?" Survey in a handful of targets, tie the photogrammetric block to them, and your deliverable inherits their accuracy. But GCPs cost time in the field, and on some sites they're impractical to place at all. The question I keep coming back to is narrower and more useful: when is network RTK alone good enough to skip ground control?

A recent 66-acre rural mapping mission gave me a clean test case. No GCPs were deployed — the aircraft logged a network-RTK Fixed solution for every image, and that was the only source of georeferencing. Here is what held up, and where the approach has hard limits.

The setup

A DJI Matrice 4E flew a nadir grid at 390 ft AGL with 80% forward and 70% side overlap, 270 images, under diffuse overcast light. Corrections came from a Point One Navigation network RTK service — no base station of my own, no surveyed targets. Processing ran through PIX4Dmatic, with DTM, contour, and zonal-statistics work finished in ArcGIS Pro.

100% camera calibration · 0.111 ft/px GSD · 33.9M dense points · geolocation RMS Z ≈ 0.116 ft — with zero ground control points.

What network RTK got right

For a property-scale orthomosaic and surface model, the RTK-only block was solid. Every image calibrated, the vertical geolocation RMS came in around a tenth of a foot, and the relative geometry across the parcel was internally consistent enough to generate 5-ft contours and parcel-clipped elevation statistics. For boundary context, volume estimates, and visual documentation, the absence of GCPs simply did not show up in the deliverable.

The key enabler is that network RTK gives you an absolute position for every camera, not just a relative one. With a strong Fixed solution and good overlap, the bundle adjustment has real-world coordinates to anchor to at every exposure — which is exactly the job GCPs would otherwise do.

Where it broke down: the canopy problem

The limit wasn't horizontal accuracy — it was ground classification under dense tree canopy. Photogrammetry only reconstructs what the camera can see, and under closed canopy the camera never sees the actual ground. No RTK solution fixes a line-of-sight problem. I had to iterate ground classification from a 5-ft DEM resolution down to 1 ft to tease apart bare earth from low vegetation, and even then the canopy areas are the weakest part of the model.

RTK fixes where the camera is. It does nothing for what the camera can't see.

So when do you still need GCPs?

From this mission and others, my decision rule:

The takeaway

Network RTK has quietly made GCP-free mapping viable for a whole class of property-scale work. The skill now is less "always place control" and more knowing the conditions under which you can responsibly skip it — and being honest about the canopy areas where no amount of georeferencing will save you.

Liquid Sun Creative · More field notes