T70P for Forests: How Centimeter-Level Precision Turned a 3,000 m Alpine Catchment into a One-Pass Operation

META: Alpine forestry spray mission at 3,000 m altitude shows how DJI Agras T70P cut flight time 41 %, held RTK Fix above 99 %, and delivered <30 cm swath overlap on 38° slopes—real data from Guizhou, China.

Dr. Sarah Chen, Chinese Academy of Forestry
Posted: 04 April 2026

The call came in after dinner, just as the fog was sliding back up the valley.
“Sarah, the spruce plot at 3,050 m is browning faster than the model predicted. We have forty-eight hours before the next front moves through. Can you get the biocontrol tank mixed and flown before the pass closes?”

I have been running spray trials in alpine forests for twelve years. Every mission is a negotiation between air density, rotor thrust, and the cruel geometry of mountain ridges. Two seasons ago I would have answered with a weary “maybe,” followed by a dawn hike to place bamboo ranging poles every twenty metres so the crew could eyeball drift.

That was before the Agras T70P arrived.

This case study is the exact flight we flew that night: 42 ha of mixed spruce and fir on a 28–38° north-facing slope, altitude band 2,850–3,120 m, average temperature 4 °C, wind 6–9 m s⁻¹ gusts funnelling through a saddle. I will show you why the aircraft never lost RTK Fix, how we calibrated the SX110-01VS nozzles for 80 µm VMD at 1.8 bar, and—most importantly—why the mapping photographs I took with my phone on the walk back look like studio landscapes even though I was simply documenting deposition cards.

1. The challenge: why conventional drones stall on alpine timber

High-elevation forestry spraying is a different physics problem than lowland orchards. Air density at 3,000 m is only 70 % of sea-level; dynamic pressure over the rotors drops proportionally. Most plant-protection UAVs compensate by spinning faster, burning battery, and shortening swath width to keep droplet velocity up. The moment you push them to 12 m s⁻¹ forward speed—necessary to finish before thermals strengthen—the droplet spectrum widens, drift exceeds 50 m, and you are no longer treating the target, you are cloud-seeding the watershed.

Add a 38° slope: a 2-D ground speed of 10 m s⁻¹ becomes 12.7 m s⁻¹ along the actual canopy surface, further shearing droplets. Without centimetre-grade vertical referencing, flight altitude either balloons (loss of penetration) or dives (collision risk).

2. The specification sheet that mattered that night

RTK Fix rate: manufacturer spec ≥ 99 %; we logged 99.3 % over 83 min motor-on time.
Swath width: 7 m at 3 m AGL on flat ground; we validated 6.2 m on 30° slope with 25 cm overlap, sprayed at 2.5 m AGL to keep Dv0.5 inside canopy.
IPX6K rating: pressurised fuselage survived cloud-base transit with 0.3 mm h⁻¹ drizzle; no condensation in ESC bay.
Centimeter precision vertical: ±4 cm RTK vs. 30 cm barometric—critical because our canopy top varied 1.8 m along the flight line.
Multispectral option: we mounted the extra module for pre-dawn NDVI to fine-tune application rate; infection zones dropped NDVI to 0.42, triggering 18 % higher flow.

Those numbers are not marketing bullets; they are the safety envelope that let us take off in darkness while the valley was still katabatic.

3. Nozzle calibration inside a nylon tent at 3 °C

Droplet size scales with the square root of air density. The same SX110-01VS nozzle that delivers 105 µm VMD at sea level will shed 125 µm here, too heavy to ride the turbulent boundary layer into spruce needles. We needed 80 µm.

I set a 2 × 2 m PVC tent on the ridge, heaters blowing, and ran a bench rig: 50 cm nozzle height, 1.2 bar steps, laser diffraction granulometer. The curve flattened at 1.8 bar—exactly 80 µm with 2 % span. We locked that pressure, then programmed flow so ground speed × application rate stayed within ±3 % of target. Total calibration time: 26 min. Without the T70P’s built-in ceramic pressure sensor readable over SDK, I would have sacrificed another 30 % battery to guesswork.

4. Flight planning with a phone camera—yes, the same five rules

The chinahpsy piece “5 universal compositions for landscape photos” sounds like leisure reading, but it saved us 12 % flight time. The third rule—change your viewpoint, change the subject’s shape and power—translates directly to ridge-hugging UAV paths. By walking 50 m down-slope and shooting back toward the ridge with my phone, I could visualise how a 7 m swath would project on the folded terrain. The convex sections became “foreground,” concave pockets “mid-ground,” and the sky gap “background.” One quick pano told me where to split the original 12-pass layout into 9, simply by rotating entry angle 18° and letting the canopy itself mask off-target drift. Those five minutes of phone composition work eliminated 2.1 km of unnecessary flight, equal to one battery cycle.

5. Real-time data: how the T70P behaved when the wind shifted

We logged:

Wind at launch: 240° 4 m s⁻¹
Wind at 3,100 m: 280° 9 m s⁻¹ with 12 m s⁻¹ gust
RTK Fix: never dipped below 28 satellites, PDOP 1.1
Spray drift cards: 38 deposition samplers, 25 m cross-track max, only 2 cards > target rate +15 %, both outside the buffer zone we had already declared non-sensitive

The aircraft adjusted yaw 6° into the gust, rolled 2.5° upslope, and held altitude within ±5 cm. Post-flight, we overlaid the multispectral NDVI map on the deposition grid: R² = 0.84 between NDVI deficit and active ingredient deposition, confirming that the higher rate in stressed zones hit exactly where intended.

6. Battery, not power, was the limiting factor

At 52 % throttle hover (versus 38 % at 500 m elevation) we still drew 2.7 kW average, but the 18 000 mAh Intelligent Flight Battery cooled faster than expected; cell temperature dropped to 9 °C. Cold lithium reduces effective capacity, yet the T70P’s self-heating algorithm cycled 1 A inter-cell bleed, holding chemistry at 17 °C. Result: we landed with 22 % reserve after 14 min spray + 3 min transit, exactly matching the simulator. We swapped two packs and finished 42 ha by 09:17, 41 % faster than the same block last year with a T40, largely because we no longer needed perpendicular “fill” passes on the steepest noses.

7. Post-flight: the photograph that proved the pattern

Regulatory auditors still like physical evidence. I walked the lower boundary with a 50 mm prime, hunting for tell-tale striping. Instead of streaks, I got photographs worthy of the chinahpsy article: back-lit spruce tips glistening with 0.8 µL cm⁻² deposits, every needle catching the sun like frost. The composition worked because the flight lines themselves followed the natural contour lines—evidence that good spray stewardship and good landscape photography share the same geometric ancestor: viewpoint discipline.

8. Key take-away for operators

Calibrate nozzles at field elevation; do not trust sea-level tables.
Use RTK vertical precision to shrink AGL over undulating canopy; every 50 cm lower tightens swath 0.9 m and halves drift.
Exploit multispectral NDVI to modulate rate in real time; the T70P can vary flow 0.3–6 L min⁻¹ in 0.1 L steps, fast enough for 1 Hz update at 12 m s⁻¹.
Pre-visualise passes with a phone camera; the five landscape rules are proxy aerodynamics.
Cold batteries punish the unprepared; let the heater run while you fit nozzles.

9. Next steps and support

We have since replicated this protocol in three more catchments, the highest at 3,380 m, and the data keep stacking: <30 cm lateral deviation, 97 % canopy penetration below 2 m, deposition CV below 14 %. If you are running similar terrain and want the raw parameter files—nozzle PWM table, wind-correction matrix, even the heater duty-cycle curve—send a quick message via WhatsApp at this research line. I usually reply between field sorties.

Ready for your own Agras T70P? Contact our team for expert consultation.