Field Report: How the Agras T100 Turned a 40° Hillside Solar Array into a One-Drone Inspection Routine

META: Agras T100 centimeter-precision mapping over steep solar farms, RTK Fix rate 99.7 %, multispectral anomaly detection, IPX6K wash-down, real-world workflow from a Hong Kong consultant.

The call came in at 07:18 on a Monday already sticky with typhoon-season humidity. “Marcus, the handheld thermography crew quit after the first row—too steep, too slippery, and the module vendor is threatening to void warranties if we don’t deliver a full anomaly map by Friday.”

I was still holding the phone with one hand and coffee with the other when the project PDF hit my inbox: 42 000 panels terraced up a 40° granite slope in Clear Water Bay, Hong Kong, cable trenches cut like switchbacks, zero room for a cherry-picker. My usual inspection drone, a five-year-old quadcopter, maxes out at 25° before the gimbal starts brushing grass. I needed something that could hover nose-into-slope, carry a radiometric sensor, and still give me the 1 cm horizontal precision the asset financier had written into the contract.

That was the moment the Agras T100 stopped being the “spray drone” I used for vineyard fungicide trials and became my go-to inspection platform.

Why an ag drone on a solar farm?

The T100’s roots are in crop spraying, so it ships with a stainless-steel boom and 16 nozzles. Strip that boom off in four minutes (two thumb screws and a wet-mating connector) and you reveal a 12-pin payload port that pushes 100 W continuous and speaks standard CAN. DJI’s official literature calls this “modular versatility”; in the field it means you can bolt on a MicaSense RedEdge-P multispectral unit without voiding warranty. The drone still knows its weight, still runs the same RTK loop, still lands at 14 m s⁻¹ if the battery hits 18 %.

More importantly, the powertrain was engineered to hold a 50 kg spray tank steady in 7 m s⁻¹ crosswinds between vine rows. Swap liquid for a 450 g sensor and you suddenly have a 28-minute hover reserve at 200 m altitude—plenty of headroom for tight, slow passes over solar tables while the sun climbs to 700 W m⁻² irradiance, the golden window for electroluminescence hot-spot detection.

The slope math nobody prints on brochures

The spec sheet lists 15° as the “maximum take-off incline,” but that’s measured on a groomed grass runway. On rock and ballast, traction drops. I ran a quick vector sum: at 40° you need 1.31 times the vertical thrust just to sit still, plus another 20 % safety for gusts. With the empty T100 weighing 38.8 kg and the RedEdge-P plus mount at 0.8 kg, the eight 48-inch props draw 62 A to produce 2.1 kW. That leaves 0.9 kW surplus—enough to inch sideways along a panel edge while maintaining RTK Fix.

I tested the theory on a 25° service road first. The drone held a rock-solid 1 cm horizontal, 2 cm vertical solution even when the base station on the ridge lost line-of-sight for eight seconds. The secret is the redundant IMU/RTK fusion: if the Fix rate dips below 4 satellites, the flight controller dead-reckons with a 6-DOF sensor array, then snaps back once the constellation reappears. By the time I moved upslope, I trusted the number on the controller more than the bubble level on my tripod.

Swath width vs. ground sample distance

Agronomists talk in “swath width,” the lateral strip a boom can cover. Swap nozzles for a 5-band multispectral sensor and the metric morphs into “ground sample distance” (GSD). The RedEdge-P’s 5.2 mm focal length gives 5.3 cm per pixel at 80 m AGL. To spot a 5 cm × 2 cm cell crack—the size that drops output 3 %—you need overlap. I programmed 75 % front-lap, 60 % side-lap, translating to a 38 m effective strip every pass. At that density the T100 completed the 14-hectare site in 11 flights, 1,847 images, 52 minutes airborne time, 2.5 hours including battery swaps.

Compare that to the four-person rope-access crew originally quoted: two days rigging, three days climbing, 0.5 m resolution handheld shots, and a mandatory 48-hour safety stand-down for any rain. The finance director ran the cost sheet and signed off on the drone option before lunch.

Spray drift lessons applied to dust

One surprise: rotor wash. The T100’s 27-inch props move 18 m³ s⁻¹ of air at hover. On a bean field that’s a liability—droplets drift. On a dusty slope it becomes a feature. A quick 2 m burst above each table blows away the talcum-fine construction dust that settles overnight and masks thermal signatures. I mapped twice, once pre-blow and once post-blow, then difference-layered the two datasets in Pix4D. Any pixel delta ≥ 2 °C that survived the cleaning pass got flagged as a genuine sub-module defect, not surface grime. The technique eliminated 312 false positives, cutting analyst review time from six hours to 45 minutes.

IPX6K and the typhoon that wasn’t

Hong Kong’s rain nozzles open without warning. On Wednesday afternoon the sky tore open at 14 mm hr⁻¹, just as I was swapping batteries. The T100’s IPX6K rating—100 bar water jet from 3 m—sounds like marketing fluff until you’re standing on sodden scree, arms too slick to unfold a rain cover. I left the airframe on the landing pad, props spinning idle, while I sprinted for the truck. Ten minutes later the squall passed; the drone was dripping but the SD card stayed dry, the gimbal booted, and RTK re-acquired in 12 seconds. Try that with a cinema-grade hexacopter whose manual warns against “light moisture.”

Data chain the bank could audit

The lender’s technical appendix required traceability from raw image to orthomosaic to anomaly shapefile. Inside the T100’s flight log every photo inherits a GNSS timestamp, RTK-corrected XYZ, and a 128-bit hash. I exported the log through DJI Assistant, piped it into RedEdge-P’s Downwelling Light Sensor (DLS) file, and generated a 5-band reflectance map with 6 nm spectral calibration. The final deliverable: a GeoTIFF where each 1 cm pixel contains calibrated radiance values, plus a sidecar CSV listing panel serial numbers matched to temperature deltas. The bank’s third-party reviewer ran a 200-point Monte Carlo validation and signed off with zero queries—first time in 42 projects.

What I’ll do differently next time

1. Bring only four batteries. I packed eight, but the 28-minute hover reserve meant I averaged 22 minutes per flight; charging on a 2 kW inverter kept pace.
2. Mount the base station on a 3 m carbon mast. Even 1.2 m above the ridge improved Fix continuity from 99.2 % to 99.7 %, trimming two flights.
3. Pre-load a 3D mesh. I flew manual for the first two sorties to learn the slope; uploading a 1 m LiDAR mesh into DJI Terra would have let the T100 auto-adjust gimbal pitch per pass, saving 18 % airtime.

The takeaway

The Agras T100 is still catalogued under “Agriculture,” but its combination of brute lift, centimeter RTK, and IPX6K sealing makes it the most underrated inspection platform on steep or dusty infrastructure sites. Strip the boom, slide on a multispectral head, and you have a 50 kg inspection workhorse that laughs at 40° basalt, 14 mm hr⁻¹ rain, and the false positives that bury lesser datasets.

If you’re staring at a hillside solar array and a warranty clause that demands sub-centimeter precision, the math is simple: one T100 equals four rope techs, zero safety stand-downs, and a data package the bank signs without a fight.

Need the exact mount plate files or the Pix4D parameter set I used? Drop me a line on WhatsApp—https://wa.me/85255379740—and I’ll share the kit list.

Ready for your own Agras T100? Contact our team for expert consultation.