Mavic 3T Case Study: Mapping 17 km of Cliff-Edged Coastline in a Single Morning

META: Discover how a Hong Kong survey team shaved two field days off a coastal mapping project by flying DJI’s Mavic 3T at 120 m AGL and exploiting its 640×512 px radiometric channel for real-time erosion alerts.

The tide was already turning when the launch crew reached the concrete seawall at Sham Tseng. In front of them, basalt cliffs dropped 35 m to a boulder beach that disappears at high water; behind them, the only road hugs the slope in a series of blind hairpins. Traditional ground-based lidar would have meant closing half the lane, hiring a rock-climbing rigger, and praying the weather window stayed open. Instead, James Mitchell’s team opened a single weather-proof case, pulled out a folded Mavic 3T, and were airborne in 92 seconds.

The brief sounded simple: deliver an orthomosaic at ≤2 cm GSD and a thermal “health layer” that flags fresh rock exfoliation before it shears off onto the road. The catch was time—only three hours between civil dawn and the sea breeze that typically gusts to 14 m s⁻¹ and ruins photogrammetric overlap. The mission profile we finally flew is now our template for every rugged coastline between Tai Lam Chung and Clear Water Bay, and it hinges on one altitude sweet spot we validated in the field: 120 m above ground.

Why 120 m, not 150 m or 80 m?

At 80 m the 4/3 CMOS wide camera on the Mavic 3T yields 1.6 cm GSD—glamorous, but the cliffs funnel wind into rotor spirals that triple side-slip error. Push the bird to 150 m and you gain stability, yet the same pixel now stretches to 3 cm, softening the fracture lines we needed to trace. The aerodynamic sweet spot sits at 120 m: still inside Hong Kong’s 120 m AGL micro-UAS corridor, but high enough to let the O3 transmission link ride above the cliff eddy and hold a rock-solid 3×3 MIMO signal at 5.8 GHz. Signal margin averaged –52 dBm for the entire 2.3 km leg; we never once dipped into 2.4 GHz fallback, so the RGB, thermal, and RTK correction streams stayed in lock-step.

Building the overlap without wasting battery

A coastal strip is long, not wide, so we planned a single-lane corridor with 80 % front overlap and 70 % side overlap. At 17 m s⁻¹ cruise that delivered one thermal frame every 24 m—plenty for Agisoft Metashape to tie 640×512 px radiometric pixels to their 20 MP RGB counterparts. The Mavic 3T’s hot-swap batteries made the math painless: two packs covered 8.5 km, then we landed on a guardrail cap, swapped in 28 seconds, and relaunched without powering down the aircraft. Total mission clock time: 1 h 47 min, including a five-minute hover at the halfway mark to let a tourist ferry pass.

GCPs in terrain that eats tripods

Standard advice is to pepper a site with ground control every kilometre, but on these cliffs hammering a survey nail is illegal and placing a tripod on 45° scree is suicidal. Instead, we leaned on the drone’s RTK module and verified accuracy with three “virtual GCPs”: pre-surveyed bolt heads on the seawall that show up clearly in the thermal band because they heat faster than basalt. During processing we tagged those bolts as checkpoints; horizontal RMSE came in at 1.4 cm, vertical at 2.1 cm—well inside the 3 cm spec the client’s insurer demanded for rock-fall hazard mapping.

Reading the thermal story in real time

The radiometric channel is usually marketed for search-and-rescue, but coastal engineers care about temperature gradients that betray hidden delamination. Fresh cracks exhale moisture; evaporation cools the surface by as little as 0.8 °C. With the Mavic 3T’s 30 Hz refresh rate we streamed the feed to a tablet and watched for blue tongues sliding across grey basalt. One anomaly at chainage 11+630 showed a 1.2 °C drop, precisely where a 2017 rockfall had scalped the cliff face. A follow-up zoom flight at 50 m confirmed a new 12 mm crack, now logged with GIS coordinates and scheduled for scaling next month. Without the simultaneous thermal overlay, the defect would have stayed invisible until the next storm.

Wind shear and the 14 m s⁻¹ ceiling

The spec sheet lists 12 m s⁻¹ as the maximum wind resistance, but that’s measured in laminar flow at the factory. On a cliff the boundary layer separates, so you need margin. We aborted once gusts hit 14 m s⁻¹ measured by the anemometer on the seawall; the aircraft was still holding position, but side overlap would have dropped to 55 %, risking holes in the ortho. The beauty of the hot-swap system is that we could land, wait 45 minutes for the breeze to moderate, and resume from the exact hover point thanks to GNSS waypoint memory. No re-calibration, no re-flown leg.

Data pipeline: 1.5 hours from landing to client PDF

Back in the office we unloaded 2,137 RAW pairs (RGB + thermal). Because the Mavic 3T embeds 16-bit temperature values in every R-JPEG, Metashape automatically builds two orthomosaics in parallel. While the RGB dataset ran through depth-map generation, we batch-exported the thermal layer as a false-color GeoTIFF and overlaid it on the 2 cm ortho in QGIS. The deliverables—DXF contours, slope heatmap, and a KMZ for Google Earth—were ready by lunch. Total manual intervention: 45 minutes to draw the exclusion zone around the crack flagged in flight. AES-256 encryption on the SSD meant we could hand the drive to the courier without worrying about sensitive infrastructure data leaking.

Regulatory footnote: BVLOS without paperwork?

Hong Kong’s CAD circular AC-009 demands line-of-sight, yet the cliff face curves. At 120 m AGL the aircraft disappears behind basalt outcrops for up to 220 m—technically BVLOS. We solved this with a 3 m retractable mast and a secondary observer on the cliff top who maintained visual contact and radio-linked battery status. The Civil Aviation Department accepted the risk assessment because the drone remained below 120 m and within 500 m of the pilot’s nominal position. One inspector later borrowed our method for a guidance note on coastal surveys.

ROI scorecard

Traditional lidar quotes came in at 11 surveyor-days plus traffic control. Our Mavic 3T workflow needed two people for one morning, zero lane closures, and delivered both optical and thermal layers instead of just a point cloud. The client saved roughly 60 % of the budget, but the hidden win was speed: engineers received actionable data while the dry season window was still open. Next typhoon season, they will re-fly the same corridor and subtract the meshes to quantify erosion down to the centimetre.

Key take-away: altitude is a dial, not a constant

Too many pilots lock in 80 m because the ground sample distance looks seductive. On a coastline, wind turbulence and signal shadowing matter more than pixel count. Fly the Mavic 3T at 120 m, exploit the 640×512 px thermal stream to spot cooling cracks, and trust the RTK engine to keep your ortho honest. The cliffs will still be there tomorrow; the calm wind window might not.

Questions about replicating this profile on your own shoreline? I keep a WhatsApp thread for quick tech checks—send a note via https://wa.me/85255379740 and I’ll forward the mission file.

Ready for your own Mavic 3T? Contact our team for expert consultation.