2026-04-25 · 9 min read
Drone photogrammetry accuracy is a topic where marketing happily oversells. The honest answer depends on three numbers — GSD, ground-control-point distribution, and checkpoint residual — and one question: how is the deliverable being audited?
GSD — the underlying ruler
Ground Sample Distance (GSD) is the size of one image pixel projected onto the ground. It is determined by camera sensor, focal length and altitude above ground level (AGL). For a typical DJI M300 + 24 mm payload at 100 m AGL the GSD is around 2.5 cm/px. At 60 m AGL it is closer to 1.5 cm/px.
GSD is not accuracy — it is the resolution of the underlying ruler. Accuracy is what fraction of one GSD pixel your final position estimate sits within.
The 1× GSD heuristic
With well-distributed RTK Ground Control Points and a properly run photogrammetry processing job (PIX4Dmatic / PIX4Dmapper / Pix4Dcloud), the rule-of-thumb absolute accuracy is:
- Horizontal: 1× GSD against checkpoints.
- Vertical: 1–3× GSD against checkpoints.
So a flight at 2.5 cm/px GSD with proper RTK GCPs gives you ≈ 2.5 cm horizontal, 2.5–7.5 cm vertical. That is the survey-grade range we quote on every drone-services engagement.
What "well-distributed RTK GCPs" actually means
A common mistake — sometimes deliberate — is to claim 1× GSD accuracy without enough GCPs to validate it. Easepect's standard for survey-grade work:
- At least 5 GCPs on a sub-10 ha site, distributed across all four quadrants and the centre.
- 1 GCP per 2–3 ha on larger sites, with denser distribution at edges.
- Independent checkpoints (not used in the bundle adjustment) for validation. We aim for at least 30% of total ground points held back as checkpoints.
- Each GCP captured by an Emlid Reach rover with a fixed RTK solution from SiReNT corrections.
Checkpoint residual — the actual accuracy number
The number that defines accuracy is the RMSE on independent checkpoints — points that were measured by RTK but were not fed into the bundle adjustment. After processing, PIX4D reports the model-vs-checkpoint residual. That residual is the real accuracy of the deliverable.
Every Easepect deliverable ships with the residuals report. If a contractor delivers a drone survey without one, treat the accuracy claim with caution.
Without GCPs — what you actually get
If you fly a drone and only use the RTK / GNSS metadata stamped on each image (no ground control), the absolute accuracy is bounded by the airframe's GNSS quality. For a non-RTK consumer drone: 1–3 m horizontal, 3–10 m vertical. For a DJI M300 RTK with PPK: typically 5–15 cm horizontal, 10–30 cm vertical — depending on satellite geometry and how strong the RTK link was during the flight.
Without RTK GCPs you are at best sub-decimetre and that's unusual. For survey-grade work, RTK GCPs on the ground are the only honest path.
When sub-cm accuracy is achievable
Sub-cm absolute accuracy is genuinely achievable on drone photogrammetry — but only by stacking specific conditions:
- Low altitude flight (30–50 m AGL).
- High image overlap (80% / 80%).
- RTK GCPs distributed every 1–2 ha.
- Independent checkpoint validation showing <1 cm RMSE.
- Calibrated camera with documented internal parameters.
This is the regime in which Easepect produces sub-cm deliverables for cadastral support, industrial precision and BIM-grade as-builts.
Terrestrial PIX4Dcatch
For terrestrial scans — building façades, equipment rooms, open trenches — PIX4Dcatch on an iPhone Pro plus an Emlid Reach RX rover delivers sub-5 cm absolute on the captured 3D model. For local feature accuracy (pipe diameter, joint position) it is sub-centimetre. This is the methodology behind the QL-A subsurface utility records trusted by the Singapore Land Authority.
Singapore-specific factors
- Multipath in built-up areas — RTK fix can be slow near tall buildings; Easepect plans GCP captures at building offsets to ensure clean sky-view.
- Heavy rain / overcast — affects image quality more than RTK; we re-fly on sunny days where photogrammetry quality matters.
- SVY21 transform — applied at output stage; covered in our SVY21 explainer.
- Surveyor sign-off — for cadastral / boundary work, Easepect partners with a Registered Surveyor for the final certificate.
Bottom line
1–3 cm horizontal, 2–5 cm vertical with RTK GCPs is the realistic survey-grade range for drone photogrammetry in Singapore. Sub-cm is achievable with extra care. Without GCPs, treat any cm-grade claim with caution — and ask for the checkpoint residuals report.
Easepect quotes accuracy as a number with a checkpoint-validated residual, not a marketing slogan. Get a quote.