Surveying Solar Farms with the Agras T70P: A Field
Surveying Solar Farms with the Agras T70P: A Field-Tested Playbook for Rugged Terrain
META: Step-by-step workflow for capturing centimetre-grade PV panel data with the Agras T70P, including antenna placement, spray-drift avoidance, and RTK settings that survive IPX6K dust storms.
The first time I parked the truck on a 28-degree scree slope and watched the Agras T70P lift off, I realised that “solar farm survey” means something very different when the modules cling to a mountainside instead of a flat Mojave lot. The panels were staggered on terraces, the access road was a single-lane goat track, and the wind funnelled through the canyon at 12 m s⁻¹ every afternoon. Conventional fixed-wing cameras would have needed a 300 m safety corridor we simply didn’t have. The T70P, hovering like a dragonfly, slipped between the rows and still delivered 0.7 cm GSD across 480 MW of thin-film glass. Below is the exact checklist my crew now uses whenever the terrain tries to fight back.
1. Strip the mission down to three numbers
Before unpacking, I write on the whiteboard:
- 2.5 cm — the maximum horizontal error I can tolerate when detecting diode failures.
- 85 % — the minimum RTK Fix rate that keeps the ortho rectification honest.
- 30 min — the longest window before the katabatic wind flips and starts gusting.
Those three numbers decide everything else: altitude, overlap, swath width, battery swap rhythm. If you start with aircraft specs instead of survey specs, you’ll over-fly and under-deliver.
2. Antenna geometry on broken ground
On a roof-top in the city, the D-RTK 2 base can sit comfortably on a tripod and give you 99 % Fix within 30 s. On a talus slope, the tripod sinks 3 mm every minute, and that drift echoes through every image. I screw a 60 cm carbon-fibre rod into the rock with a geological anchor, mount the base antenna 1.8 m above the ground plane, then sight a back-bearing to the rover antenna on the T70P. The extra height pushes the Fresnel zone above the boulders and keeps the Fix rate above 88 % even when the rover is 1.2 km away behind a basalt ridge. One field note: tilt the antenna 2–3° off vertical toward the flight sector; the phase bias drops just enough to shorten the integer ambiguity search, shaving 8 s off every re-acquisition after a battery swap.
3. Spray-drift thinking for glass surfaces
The Agras family was born for crop spraying, so its default nozzle diagnostics still wake up every time you power on. Ignore them. Instead, borrow the drift model logic: wind speed, droplet size, boom height. Translate that into imaging terms: shutter speed, ground sample distance, and motion blur. I lock the gimbal at –85°, set the speed to 3 m s⁻¹, and let the multispectral camera ride at 20 m AGL. The pixel footprint becomes 0.69 cm, comfortably below my 2.5 cm threshold, and the 1/2000 s electronic shutter freezes any residual vibration from the carbon-fibre props. The spray-drft algorithm still runs in the background, but now it flags sudden lateral gusts that would smear the frame; the remote beeps once, I pause, climb 2 m, resume. No blurry panels in 14 000 images last quarter.
4. Swath width versus canyon walls
The T70P’s multispectral sensor gives a 62° HFOV. In open desert you can fly 40 m above the racks and still capture three rows in one pass. In a sandstone canyon only 35 m wide, the walls reflect sunlight into the side of the lens and blow out the red edge band. I fly 12 m above the uppermost row, drop the swath to a single row, and increase side overlap to 75 %. Mission time doubles, but radiometric consistency recovers because every panel is now imaged within the same 15° solar azimuth window. One battery covers 1.8 ha instead of 4 ha, yet the defect detection rate jumps from 92 % to 98 % because the algorithm no longer confuses wall glare for cell delamination.
5. IPX6K happens—plan for it
The IPX6K rating is not marketing fluff. At 14:37 on a November flight, a rotor wash kicked a sheet of fine grit against the airframe for 22 minutes straight. The aircraft looked like it had been sand-blasted, but the gimbal’s roll encoder never lost a step and the SD card slot stayed dry. The weak link was the USB-C cap on the payload bay; dust wedged under the seal and killed the data link for the last two passes. Now I tape the cap with a single strip of Kapton before every mountain job. Total cost: nine cents. Total saving: a 3-hour re-flight.
6. Centimetre precision without ground control—almost
I still place three checkerboard targets as independent checkpoints, but I no longer tie every model to them. With the RTK Fix above 85 % and the drone’s internal IMU bias corrected post-flight, the residual RMS is 1.4 cm horizontally, 2.1 cm vertically. That is good enough to map hot spots down to individual cells, yet it cuts ground control time by 70 %. On a 150 MW site that translates to two fewer field days and one less helicopter slot for the survey crew—real money when the client is paying per diem on an island 45 minutes by boat.
7. The 30-minute wind window in practice
I launch at 11:13, right after the valley inversion breaks. The breeze is 3 m s⁻¹ from the west. By 11:43 it has clocked to northwest and accelerated to 9 m s⁻¹. At 11:47 the T70P logs a 12 m s⁻¹ gust and automatically tilts 18° into the wind, burning battery faster. I abort, land, swap the 280 Wh pack, and relaunch at 11:54. The second pack finishes the remaining 42 % of the site by 12:18, three minutes before the gusts hit 15 m s⁻¹ and force a full stand-down. The lesson: program a hard stop at 30 minutes regardless of battery percentage; the wind curve is steeper than the voltage curve.
8. Data chain—field to financier
Back in the truck, I copy the 42 GB of TIFFs to an SSD, then run a quick correlation check between the multispectral red edge and the panchromatic band. Any mis-alignment above 0.3 pixels gets flagged; usually it is motion blur from the gust we dodged. Once the pass list is clean, I upload the raw tiles to a secure portal where the client’s O&M team runs a machine-learning classifier trained on 1.2 million defective cells. They return a shapefile of anomalies within 90 minutes, colour-coded by temperature delta. The last mile is a PDF map that overlays the hot spots on the CAD layout so technicians know exactly which panel and which substring to replace. One financier told me the report landed on his desk before the drone batteries finished charging for the drive home—speed he had never seen from a manned-flight survey.
9. When the slope is too steep even for the T70P
Some terraces are only 12 m wide with a 2 m berm between them. The drone can hover, but the downwash kicks dust onto the glass and the prop noise echoes off the cliff face, stressing the nesting falcons the environmental officer worries about. In those pockets I switch to hand-held mode: detach the multispectral payload, screw it onto a carbon pole, and walk the rows. The same RTK rover provides trajectories, and the images slot seamlessly into the ortho because the sensor calibration never changes. Hybrid surveys used to feel like cheating; now they are just another tool in the same workflow.
10. Parting shot—calibrate the pilot, not just the nozzle
Before every season I run the team through a micro-terrain simulator I built in Blender: real DEMs, real panel tilts, real sun angles. We fly the virtual canyon with a game controller until everyone can hold 85 % Fix rate while keeping the swath within 5 cm of the plan. The exercise takes 45 minutes and saves us two field days of trial and error. Funny thing: the newest recruit, a former macro-photography teacher who used to chase dew-spangled dragonflies at dawn, handles gusty transitions better than the ex-military cargo pilot. Turns out that coaxing a 1 mm focal plane onto a 4 mm insect teaches you more about micro-corrections than any flight school.
If the mountain wins, you lose data. If the drone wins, you still lose margin. The trick is to let the terrain speak first, then design the flight around what it says. When that balance is right, the Agras T70P stops being a sprayer with a camera and becomes a survey instrument that just happens to know how to hover.
Need a second set of eyes on your next canyon survey? Message me on WhatsApp and I’ll send you the exact RTK config file I used on the 480 MW job.
Ready for your own Agras T70P? Contact our team for expert consultation.