UAV Photogrammetry for a 1:500 Site Plan: Feasibility and Vertical Accuracy
A 0.54 km² site (900 m × 600 m) comprising a gravel pit and surrounding scrub must be mapped for a 1:500 plan with a contract limit of RMSEz ≤ 0.040 m. The UAV carries a 20 MP camera (5472 × 3648 px, 2.4 µm pixel pitch, 24 mm lens) flown at 120 m AGL with 80% forward overlap and 60% sidelap. The exposure interval is 2 s and a battery yields 22 minutes of usable flight. Fourteen ground control points were pre-marked: nine on the open pit floor, five in scrub. Thirty independent checkpoints produced RMSE horizontal 0.035 m and RMSE vertical 0.058 m. Checkpoints in the pit show a mean vertical residual of +0.012 m; those in scrub show +0.108 m. Deliverables are a dense point cloud, an orthomosaic and 0.10 m contours.
1. Verify the flight design: ground sample distance, image footprint, number of images and flight time. Is the mission feasible as planned?
Read the worked answer
Geometry first.
- Sensor: 5472 × 2.4 µm = 13.13 mm wide; 3648 × 2.4 µm = 8.76 mm high.
- GSD = H × pitch / f = 120 × 0.0024 / 24 mm = 0.012 m (12 mm) - far finer than a 1:500 plan needs.
- Footprint = 120 × 13.13 / 24 = 65.7 m across-track; 120 × 8.76 / 24 = 43.8 m along-track.
- Sidelap 60% → line spacing = 0.40 × 65.7 = 26.3 m → 600 / 26.3 ≈ 24 lines.
- Overlap 80% → exposure spacing = 0.20 × 43.8 = 8.76 m → 900 / 8.76 ≈ 104 images per line.
- Total ≈ 24 × 104 ≈ 2,500 images.
Time. At a 2 s interval the aircraft must fly 8.76 / 2 = 4.4 m/s. One line takes 900 / 4.4 ≈ 205 s, so 24 lines ≈ 82 minutes of straight-line flying before turns. At 22 minutes per battery that is four battery changes minimum, realistically five with turns and re-flights.
Judgement. Feasible in a single day but not in a single flight, and the plan is wasteful: 12 mm GSD is roughly four times finer than required. Flying at 180 m AGL gives 18 mm GSD, 16 lines and about 1,100 images - under 45 minutes, two batteries, with accuracy unchanged. The finer GSD buys nothing and costs processing time and battery logistics.
2. Does the delivered product meet the accuracy specification? Decompose the error to identify what is actually wrong.
Read the worked answer
No - RMSEz is 0.058 m against a 0.040 m limit, a 45% overshoot.
The aggregate number is not the useful part; the split is:
| Zone | Mean vertical residual |
|---|---|
| Pit (open, 9 GCPs nearby) | +0.012 m |
| Scrub (5 GCPs, vegetated) | +0.108 m |
Two conclusions follow.
1. The error is spatial, not random. If the block were misregistered overall, pit and scrub would share the same offset. Instead the residual grows exactly where the control is sparse and the surface is vegetated. With 9 of 14 GCPs in the pit, the bundle adjustment is extrapolating into the scrub - a 0.108 m extrapolation over roughly a third of the site is entirely expected.
2. The surface in scrub is a DSM, not a DTM. Dense image matching cannot see the ground through scrub; it returns the top of the canopy. The +0.108 m is partly real canopy height, and no amount of adjustment removes it. Contours drawn from that surface will float above the ground in exactly the areas where earthworks are least visible.
Supporting check: horizontal RMSE of 0.035 m is acceptable for 1:500, which reinforces that the problem is vertical/classification rather than a gross georeferencing failure.
3. Specify the re-flight and processing changes you would require before accepting the deliverable.
Read the worked answer
Fix control distribution, then fix the surface.
Control and georeferencing:
- Redistribute GCPs to be proportional to land cover: at least 5 in scrub, well spread, with no GCP-free zone larger than about 150 m.
- Add redundant checkpoints in scrub as the acceptance test; the current 30 checkpoints mask the problem by being dominated by open ground.
- Add PPK/RTK camera positioning to reduce dependence on GCP density, but keep GCPs as the independent check.
Flying:
- Raise altitude to ~180 m (18 mm GSD) - accuracy is limited by control and surface classification, not pixel size.
- Add cross-strips over the scrub and raise sidelap there; add a low oblique pass to break up matching failure in uniform vegetation.
- Fly nearer midday-flat light or with stable cloud to reduce shadow-induced matching bias in scrub.
Processing:
- Classify ground returns and enforce a ground-class filter in scrub, with manual breaklines along vegetation edges and pit crests.
- Deliver a DTM, not the raw dense-cloud surface, and state the method used.
- Re-run camera self-calibration with the new control geometry and verify focal length against the two-peg-style check of known distances.
Acceptance: re-issue RMSEz and a residual-vs-land-cover table; accept only at RMSEz ≤ 0.040 m with no land-cover class mean residual above ±0.020 m.