Do You Still Need Ground Control Points with RTK Drones?

On This Page

• What RTK, GCPs, and checkpoints each do
• When to retain ground control
• What no-GCP research actually demonstrates
• Plan the fieldwork before reducing targets
• Read the accuracy report correctly
• Questions to put in the supplier brief

What RTK, GCPs, and checkpoints each do

Real-time kinematic positioning, or RTK, uses satellite observations and corrections from a base station or reference network to improve the positions recorded during image capture. Photogrammetry software combines those positions with overlapping images to reconstruct the scene. Post-processed kinematic positioning, or PPK, applies corrections after capture and requires suitable recorded observations and a supported processing workflow. Neither method directly measures every point on the finished surface.

A ground control point is a recognizable location with surveyed coordinates that the processing software uses to constrain the model. A checkpoint has surveyed coordinates held out of that adjustment; comparing its reconstructed and surveyed positions tests the result. The physical target can look identical. Its assigned role in processing determines whether it is control or a check. See PIX4D's explanation of control and check points.

Consequently, a proposal for “zero GCPs” can still include substantial ground measurement. Ask how many points constrain the model and how many remain reserved for assessment. A small residual at a point the software was told to fit is not an independent test of the surrounding map.

Also separate relative accuracy, the agreement of dimensions and positions within a model, from absolute accuracy, its agreement with the required coordinate reference frame. A model can preserve useful internal dimensions while sitting in the wrong location. PIX4D's accuracy guidance distinguishes these properties and explains why image quality and reconstruction still limit an RTK dataset.

When to retain ground control

Choose the control approach from the required output and the conditions that could invalidate it. The following table is an editorial synthesis of the technical guidance and field studies cited below, checked September 7, 2026. It is a scoping aid, not a prescribed target count.

GCPs cannot repair missing image detail. Blur, occlusion, poor overlap, and difficult surfaces can impair reconstruction even when positioning is good. Likewise, ground sampling distance, the ground length represented by a pixel, is a resolution measure rather than proof of positional accuracy. The USGS calibration guidelines connect geometric quality to control, camera calibration, tie points, and acquisition geometry.

Before adding control to correct a consistent height offset, confirm the height reference. Ellipsoidal heights and orthometric heights refer to different surfaces. Image positions, surveyed points, and exported deliverables need compatible coordinate definitions and the appropriate transformation. PIX4D's geoid guidance explains why a constant height shift is only suitable under limited conditions. Check the capabilities of the particular software and version being used.

What no-GCP research actually demonstrates

There is credible evidence that GCP-free RTK mapping can work. In a 2020 River Feshie study by Stott, Williams, and Hoey, researchers flew a DJI Phantom 4 RTK over a roughly 2 by 0.5 km river reach. They used a double grid with imagery angled 20 degrees from nadir and assessed the outputs against 3,300 spatially distributed RTK-GNSS checkpoints.

The no-GCP scenario produced a vertical root mean square error of 0.066 m, compared with 0.072 m for the five-GCP scenario. Neither scenario showed the systematic vertical doming discussed in some other investigations. Those results demonstrate a successful configuration under the study's conditions. They do not establish that removing GCPs improves accuracy generally, or that another aircraft, camera, site, and flight plan will reproduce the result.

Other research helps explain why outcomes vary. A 2021 study by Obanawa, Hayakawa, and Sakanoue examined RTK surveys without GCPs and found that camera angle, camera-position accuracy settings, and lens-calibration treatment affected doming and vertical error. Doming is a broad curvature error in the reconstructed surface; accurately positioned photographs do not automatically rule it out.

The purchasing implication is to request a representative trial with independent ground checks. A study can establish feasibility. Your own delivery requirements determine whether its errors would be acceptable for your job.

Plan the fieldwork before reducing targets

Start with the recipient's intended use: an orthomosaic for visual context, a surface for earthwork measurements, or data that must align with an engineering model. Agree the horizontal and vertical error measures, units, required coverage, coordinate system, height reference, and any named standard and edition. Set the point-count and distribution requirements before capture rather than choosing them after seeing the result.

Then assemble the inputs needed to produce and check that output:

1. Positioning records: identify the RTK correction source, its coordinate reference, and the basis for any local base coordinates. Retain the image geotags and available positioning-quality records. If PPK is a contingency, confirm that the aircraft records the necessary observations and exposure information before flying.
2. Ground measurements: establish suitable surveyed points, record coordinate units and measurement uncertainty, and assign control and check roles. Plan recognizable, stable targets across the area and elevation range that matters to the deliverable. Keep a record of target IDs and field observations.
3. Capture and processing settings: document the camera, flight geometry, overlap, software version, calibration approach, and treatment of image-position uncertainty. Preserve the images and settings needed to reproduce the output.

These are proposed handover requirements drawn from the cited positioning and calibration guidance. Their purpose is to make the service reproducible and to expose gaps before the crew leaves the site. Also ask how the reference survey was verified: as a matter of measurement logic, if the drone and checkpoint survey inherit the same wrong base coordinate or transformation, their agreement alone cannot reveal that shared mistake.

After capture, inspect the images and correction-quality records before committing to a no-GCP solution. An “RTK-equipped” aircraft label does not show whether all exposures received suitable corrections. PIX4D's published RTK/PPK field comparison describes unevenly distributed float positions during interrupted corrections and the resulting local accuracy problems. Its site-specific results are a reason to examine where degraded positioning occurred, rather than accepting one flight-wide percentage.

Process with the intended control set and evaluate the held-out points. If a failed checkpoint is used to adjust the model, it becomes control for that version. Preserve other independent checks or collect additional measurements before claiming the revised model passes. Record rejected measurements and the reason for rejection; do not simply remove the largest errors to improve the report.

Read the accuracy report correctly

Require separate results for the GCPs used in adjustment and the checkpoints used for assessment. The checkpoint table should identify each point, its surveyed and reconstructed coordinates, signed differences, and whether it was included in the final statistics.

For vertical differences defined as model height minus surveyed height, the basic calculations are:

Mean vertical error = sum of vertical differences / number of checkpoints.

Vertical RMSE = square root of the mean of the squared vertical differences.

Mean error exposes a common shift. RMSE reflects error magnitude, including both bias and scatter. Neither is the maximum error, and neither establishes a confidence level without the relevant statistical method.

For illustration only, suppose four checkpoint differences are +0.02, -0.02, +0.04, and -0.04 m. Their mean is 0 m, while their vertical RMSE is sqrt((0.0004 + 0.0004 + 0.0016 + 0.0016) / 4) = approximately 0.032 m, rounded to the nearest millimeter. The errors cancel in the mean while remaining present in the surface. These hypothetical inputs are not a flight result or a recommended sample size.

Ask for a map of the checkpoint differences as well as summary statistics. A few good checks near the launch area do not establish accuracy across an untested corridor or hillside. The River Feshie researchers specifically recommend spatially distributed independent checkpoints to detect systematic errors.

Finally, verify the product actually being delivered. A reconstructed checkpoint report and a finished ground-surface check answer related but different questions. If the order specifies a terrain surface, agree how its elevations will be compared with suitable ground measurements, including areas where vegetation or reconstruction artifacts complicate interpretation. USGS describes both control-based calibration and checks on data quality; our explanation of what drone inspection evidence can establish develops the broader distinction between an image, a measurement, and a supported conclusion.

Questions to put in the supplier brief

Ask the supplier to answer these before quoting a reduced-control workflow:

• Which points will adjust the model, and which will remain independent checks?
• What coordinate system, height reference, units, and transformations will the deliverables use?
• Which horizontal and vertical statistics will determine acceptance, and how will the checkpoint survey's uncertainty be reported?
• How will checks cover the delivered area, terrain types, and elevation range?
• What happens if corrections fail or the first output misses the agreed error limits? Who supplies the additional measurements or reflight?
• Will delivery include the point-role list, per-point errors, spatial error map, processing report, and the requested orthomosaic or surface files?

Compare proposals on that complete scope. A quotation with fewer adjustment targets may still include checkpoint surveying, correction services, processing, and recovery work. No general savings percentage follows from choosing RTK.

Use RTK to reduce ground control where a representative workflow and independent checks justify it. Keep GCPs where they are required or demonstrably useful, and make verification part of the job from the beginning. The useful commercial outcome is an accepted mapping deliverable with known limits.

Source notes

• PIX4D: RTK and PPK workflow. Software-provider documentation on correction methods, processing inputs, and checking RTK/PPK results.
• PIX4D: Tie points in a photogrammetry project. Technical documentation distinguishing adjustment control from checkpoints.
• PIX4D: Relative and absolute accuracy of drone mapping. Definitions and reconstruction limits; expected accuracy is conditional.
• USGS: Guidelines for Calibration of Uncrewed Aircraft Systems Imagery. Government technical report, 2023-1033, covering acquisition, control, calibration, and data-quality checks.
• PIX4D: Outputs with respect to a geoid model. Height-reference principles and conversion limitations; software capabilities differ by product.
• Stott, Williams, and Hoey: Ground control point distribution for accurate kilometre-scale topographic mapping. Peer-reviewed 2020 field study of one RTK aircraft and river survey configuration.
• Obanawa, Hayakawa, and Sakanoue: Methods to reduce doming without GCPs in RTK-UAV surveys. Research paper, 2021, relating image geometry and processing choices to vertical error.
• PIX4D: RTK vs PPK drones vs GCPs. Provider's field comparison, updated January 31, 2022, illustrating correction interruptions and spatial variation in error.

Last checked: September 7, 2026.