01The challenge
Tension cracks had opened in the verge and nearside lane of a single-carriageway A-road running along the crest of a clay embankment. The council's highways team had a single inclinometer installed mid-slope, but it could only report movement at one point, and the embankment face was covered in mature scrub and trees that made a conventional walked survey slow and hazardous.
The team needed to know how much of the 620 m embankment was moving, how fast, and whether movement was accelerating, so that it could decide between a targeted repair and a full-length scheme before the next winter.
- Dense vegetation on the slope face ruled out photogrammetry as the primary method.
- The carriageway had to stay open: no lane closures or temporary traffic management were available for survey.
- Results had to be repeatable to a known, documented tolerance so that genuine movement could be separated from survey noise.
02Our approach
We flew a DJI Matrice 350 RTK carrying a Zenmuse L2 LiDAR sensor from farmland on the downslope side. Flight lines ran parallel to the road at 80 m above ground and were offset so the aircraft stayed outside the required separation from the live carriageway, with the sensor's 70° field of view still covering the crest, verge and road edge. Each epoch was captured in a single morning.
| Parameter | Value |
|---|---|
| Aircraft / sensor | DJI Matrice 350 RTK / Zenmuse L2 |
| Flight height | 80 m AGL, 6 m/s, 50% side overlap |
| Returns | Five returns per pulse, repetitive scan pattern |
| Mean point density | ≈240 pts/m² all returns; ≈45 pts/m² classified ground under canopy |
| Positioning | Network RTK with local GNSS base, post-processed trajectory |
| Control | 8 permanent GCPs (survey nails and painted targets), 12 independent checkpoints |
| Coordinate system | OSGB36 (OSTN15) / ODN (OSGM15) |
Control points were installed once, in locations clear of the slip, and re-observed each visit with static GNSS to confirm they had not moved. Ground points were classified with a combination of automatic filtering and manual editing, then compared epoch-to-epoch using M3C2 point cloud distances and DTM differencing. A stable reference area on the opposite side of the road was used to derive a level of detection, so every mapped change came with a stated confidence.
03The outcome
The baseline and four follow-up epochs showed that movement was confined to a 70 m section rather than the full embankment. Over twelve months the crest in that section settled by up to 190 mm, with corresponding bulging of up to 140 mm at the toe: a classic rotational failure pattern. Elsewhere, measured change sat within the ±35 mm level of detection.
- The design of a soil-nailed repair was limited to the active 70 m section, avoiding a full-length scheme.
- Rate-of-movement plots showed acceleration after prolonged winter rainfall, which the council used to justify emergency funding.
- The inclinometer readings agreed with the LiDAR-derived movement at its location to within 12 mm, giving the team confidence in the wider dataset.
- The permanent control network means monitoring can continue after the repair to confirm it is performing.
04Deliverables
- Classified LiDAR point clouds for each epoch (LAZ, ASPRS classes)
- 1 m and 0.25 m DTMs for each epoch (GeoTIFF)
- Epoch-to-epoch change maps and M3C2 difference clouds
- Cross-sections at 10 m chainage intervals (DWG and PDF)
- Survey control report with checkpoint RMSE for every epoch
- Movement summary report with level-of-detection statement
