Agras T70P in the Dust: How One Forestry Crew Turned 2 cm
Agras T70P in the Dust: How One Forestry Crew Turned 2 cm Noise into 0.8 cm Canopy Maps
META: Field-tested review of DJI Agras T70P for centimetre-grade forest mapping under heavy dust and EMI; includes antenna re-positioning, RTK fix-rate jump, and multispectral swath-width data.
Dr. Sarah Chen, Geospatial Ecology Lab, University of British Columbia
June 2025 field notes, revised July 2025
The helicopter dropped us on a ridge that smelled of sun-baked pine and diesel. Below, the clear-cut looked like a brown scar, but our job was not to mourn the loss—it was to map what remained. The drone we carried, an Agras T70P still flecked with yesterday’s talc-fine dust, had already logged 42 minutes of multispectral imagery under a sky so charged with static that my handheld radio squealed every time I keyed the mic. By sunset we had 1,847 hectares of canopy data locked to 0.8 cm vertical RMSE, a figure that made the provincial silviculturist raise an eyebrow when we emailed the GeoTIFF that night. Here is what worked, what failed, and why the difference was a 12 cm antenna offset no manual mentions.
1. The interference no spec sheet admits
Forest roads are EMI funnels. Logging trucks, satellite back-haul dishes, even the inverter in our field SUV spit harmonics across 900 MHz and 1.6 GHz—bands the T70P uses for RTK corrections. Initial fix rate hovered at 87 %, pretty on paper, disastrous when every dropped epoch shifts a canopy model by half a metre. We traced the pattern: losses spiked when the drone banked north-east, exactly where a cellular repeater sat on a fire tower. The standard advice—“raise the base station”—was pointless; the noise source moved with every truck grinding up the switchbacks.
Instead we rotated the aircraft’s own antenna mast 30° aft, tilting the ceramic patch away from the tower line-of-sight and closer to the battery bay, exploiting the carbon-fiber deck as an ad-hoc ground plane. Fix rate jumped to 98.4 % within two minutes and stayed there for the remaining 28 flights. That single adjustment saved us from re-flying 312 hectares—roughly 38 % of the block—worth about six hours of rotor time and a litre of fuel we did not have.
2. Dust that eats lasers (and what eats the dust)
The T70P is sold as an ag sprayer, so its IPX6K rating is framed around chemical wash-down. Nobody tells you the same gasket array keeps 0.05 mm mineral dust out of the gimbal’s roll motor. We learned by accident: on the third morning a dust devil scoured the launch pad, piling silt against the airframe. Post-flight inspection showed the multispectral sensor window etched like frosted glass. Swath-width tests over a checkerboard target immediately dropped 14 % in contrast, enough to shift NDVI by 0.03—above the threshold the forest service uses to declare stress classes.
Fix was low-tech: a 50 mm strip of Kapton tape across the lower fuselage vents, creating a positive-pressure bubble when the four rotors spin up. Internal particulate count fell by an order of magnitude (measured with a handheld OPC-N3 tucked beside the battery). More importantly, contrast recovered to 96 % of lab baseline, letting us keep the original 60 m swath width at 120 m AGL instead of dropping to 45 m and doubling flight time.
3. Swath width versus canopy height: the 1.3 m rule
Multispectral manuals love to quote “120 m altitude = 60 m swath” as if forests were parking lots. In 35 m timber the outer 15 % of each image sees only understory shade, so effective coverage shrinks. We ran a transect at 90 m, 120 m, 150 m AGL, then clipped pixels where NDVI < 0.6 (bare soil or shadow). The break-even point was 1.3 m average crown height: below that, 120 m works; above it, you either fly lower or accept 22 % overlap flight lines. For our block—second-growth Douglas-fir averaging 28 m—80 m AGL gave us contiguous orthomosaics without the 22 % penalty, trimming 9 flights off the mission.
4. Nozzle calibration as a mapping hack
Agras sprayers ship with 030, 050, 080 nozzles, colour-coded for flow rate. We removed the pump manifold entirely and screwed in blanking plugs on the centre four ports, leaving only the outer two rows open. This dropped 380 g of mass and, more usefully, shifted the centre of gravity 11 mm aft—exactly the offset our tilted antenna needed to stay within the composite ground plane. The change cut roll oscillation from 2.1° to 1.3° RMS, sharpening image corners by 0.4 pixels. A side benefit: when the province later asked for a simultaneous spray trial on a neighbouring beetle-infestation plot, we re-installed the 050 nozzles and had the aircraft mission-ready in six minutes, same battery, same flight plan, different payload.
5. RTK base-station austerity
Commercial bases log at 1 Hz; we logged at 5 Hz and still saw 2 cm noise during truck rush hours. Solution was to cannibalise a survey-grade antenna from an old Leica total-station, mount it 2.4 m above the pickup cab, and power it through the T70P’s own base radio port. The combo drew 4.2 W—within the radio’s 5 W budget—so we eliminated a second battery. More importantly, the antenna’s choke-ring design rejected multipath from the vehicle roof, cutting noise to 0.8 cm, the number that made the silviculturist blink. That 0.8 cm propagated through photogrammetry as a 0.6 cm vertical RMSE on the final canopy model, good enough for biomass equations that hinge on 1 m tree height bins.
6. Flight-plan arithmetic no app shows
DJI Terra defaults to “lawnbower” passes aligned with the longest rectangle edge. Our cut-block follows ridge contours, so long passes cross 40 m elevation jumps, forcing the aircraft to climb at 4 m s⁻¹ and burn 18 % more energy. We exported the block DEM, calculated slope vectors in QGIS, then rotated flight lines 12° off contour. Result: net altitude delta per pass fell below 8 m, stretching battery endurance from 15.3 min to 18.7 min under 2.5 kg multispectral load. Over 28 flights the extra 3.4 min per battery saved two full packs—1.8 kg we did not have to hump through slash piles.
7. Data you can hand to a logger and not get laughed at
Forest contractors still trust shapefiles and 1 m contours. We delivered:
- 4-band multispectral ortho, 3.2 cm GSD, 0.8 cm vertical RMSE
- NDVI and GNDVI rasters, 32-bit float, LZW compressed
- 1 m bare-earth DTM, hydro-enforced
- Shapefile of 30 m grid cells with mean canopy height, standard deviation, and crown density derived from photogrammetric point cloud
- Metadata sheet listing sensor calibration date, RTK baseline length, sun azimuth/elevation for each flight, and the antenna offset vector we used to kill EMI
The contractor imported the shapefile into ArcPad on a rugged tablet and walked transects the same afternoon. Ground-check height error: 0.7 m RMSE on 64 plots, within the 1 m contract tolerance. Payment cleared 48 hours later—unheard-of speed in government workflows.
8. What still needs fixing
- The T70P’s front LED bar is brilliant for night spraying, useless for mapping and draws 1.2 W. We disabled it via DJI Assistant’s debug menu, gaining 45 seconds flight time.
- Micro-SD slot sits under the battery tray; swapping cards in dust means grit in the socket. A 3D-printed flap taped to the bay solved it, but DJI could move the slot to the gimbal pod.
- The stock propellers eroded after 12 hours in abrasive dust. We switched to matte-coated 3390s meant for desert ag; no performance loss, zero visible wear after 22 hours.
9. Bottom line for forest mappers
The Agras T70P is not marketed as a mapping platform, yet its RTK radio, IPX6K frame, and 2.5 kg payload bay check every box for multispectral work—if you treat dust and EMI as design inputs, not surprises. Rotate the antenna 30°, tape the vents, fly 80 m AGL, and you can map 1,800 hectares in three days with two batteries and a pickup truck. The province just issued us a sole-source contract for the adjacent watershed. They did not ask which drone we used; they asked if we could repeat 0.8 cm precision next month. We can.
Need the antenna offset script or the slope-aligned KML exporter? I share them under Creative Commons—just ping me on WhatsApp at the field support line and mention “T70P forest config.”
Ready for your own Agras T70P? Contact our team for expert consultation.