A client once asked us how much a power line weighs. Not in tonnes: in gigabytes. It is a better question than it sounds, because it forces you to walk the whole road from an aircraft with a hole in its floor to a two-page report that says “prune here, replace this conductor, leave the rest alone”. This field note walks that road with a typical project: an 80-kilometre high-voltage line over farmland and forest, with the numbers we actually work with.
Day 0: the flight plan
Everything starts with three numbers: point density, flying height and overlap between strips. The public benchmark in Spain is the PNOA-LiDAR programme run by the National Geographic Institute (IGN), which has raised the bar with every coverage: 0.5 points/m² in the first (2009–2015), 1, 2 and 4 points/m² by lot in the second (2016–2021) and 5 first-return points/m² in the third (2022–2025), with a required vertical accuracy of 10 centimetres (RMSE) and a maximum of 20 centimetres in 95% of cases. The third-coverage specification also calls for at least 15% cross-strip overlap on flat terrain, a maximum effective field of view of 50º and 60% forward overlap for the auxiliary camera.
That is country-scale mapping. A power line is a different animal. To measure the gap between a 3-centimetre conductor and the crown of a pine you need tens of points per square metre, not five, and the flight is not a block but a corridor: strips along the line axis, at low altitude, with generous overlap so that the conductor, thin and a poor reflector, shows up in more than one strip. Windy days are cancelled, not for the pilot's sake but because the conductor swings and the cloud comes out smeared.
Day 1: what goes on board
The scanner sets the pace. A RIEGL VQ-1560 II-S, to take a high-end unit with a public datasheet, combines two channels up to a 4 MHz pulse rate, achieves up to 2.66 million measurements per second on the ground, handles 45 pulses in the air simultaneously and quotes 20 millimetres of accuracy and precision at 250 metres. With those figures, flying at 1,600 metres already yields national-mapping densities; dropping to 400 metres over a corridor gives the tens of points per square metre a line needs.
But LiDAR only sees geometry. Alongside it fly an RGB camera (for the orthophoto, and so the maintenance engineer recognises what they are looking at), a multispectral camera (near infrared separates live vegetation from dead, which matters for fire risk) and a thermal one (a hot splice is a splice that is going to fail). Four sensors, four clocks, four calibrations. The least glamorous and most decisive part of the project is that all four are synchronised to the microsecond and pointing exactly where we think they are pointing.
Day 1, afternoon: knowing where the aircraft was
A LiDAR point is a range and two angles. Turning it into coordinates requires knowing, for every pulse, where the sensor was and which way it was looking. That is the job of GNSS and the inertial measurement unit (IMU). A post-processed Applanix POS AV 610 quotes 2 centimetres horizontally, 5 vertically, 0.0025º in roll and pitch and 0.005º in heading. It sounds like nothing until you multiply it out: 0.005º at 400 metres is 3.5 centimetres of ground shift from heading alone. That is why we fly with GNSS reference stations on the ground, and why we survey control fields: flat, bare surfaces (car parks, runways, roofs) where we check the cloud lands where it should. PNOA requires the block's relative geometry to stay within 3 centimetres RMSE in Z and 8 in XY. On a power-line corridor we are stricter, because the client will make felling decisions with those centimetres.
Day 2: the weight of the line
Back to the client's question. A 2020 paper on big LiDAR data does a simple sum: a 2-million-points-per-second system running for five hours produces about 36 billion points, close to 50 billion once multiple returns in vegetation are counted; at 60 bytes per point, roughly 3 terabytes of laser data alone. That is in the order of 600 gigabytes per flight hour. The same paper estimates that 10 points/m² plus a 5-centimetre orthophoto take about 0.8 gigabytes per square kilometre.
At national scale the IGN offers a tangible figure: the Asturias cloud from the second PNOA coverage (UTM zone 30) is 3,284 files and 403 gigabytes in LAS, which compress to 68.5 gigabytes in LAZ, about 17%. Our 80-kilometre line, flown low with four sensors, runs to hundreds of gigabytes: a fraction of Asturias in area, nothing trivial on disk. That is the answer: a line weighs as much as you want to know about it.
Days 3 to 10: teaching the cloud what things are
So far we have billions of points with coordinates and nothing else. Classifying means telling each one what it is. The ASPRS LAS standard reserves codes for the job: 2 is ground, 3, 4 and 5 are low, medium and high vegetation, 6 building, 9 water, 11 road surface. And from version 1.4 there are classes made for us: 13 shield wire, 14 phase conductor, 15 transmission tower, 16 wire-structure connector (the insulator). That the international point-cloud standard has a class for the insulator says a good deal about what airborne LiDAR is used for around the world.
PNOA delivers its points as class 0 and noise as class 7, then classifies with an automatic pass and a supervised refinement. On a power line the classification is finer and more thankless: the conductor is a catenary a few centimetres thick that a generic ground algorithm mistakes for noise, and the vegetation under the line is precisely what cannot be lost. That is why we run our own classification software, trained on thousands of kilometres of corridors, rather than trusting the default parameters of a general tool. A field-notebook rule: if the conductor comes out classified as “high vegetation”, the pine has not grown, the model needs checking.
Day 11: deliverables
From the classified cloud come the products, and it pays to know which does what.
- DTM and DSM: the digital terrain model (ground only) and the digital surface model (everything on top). The difference between the two is vegetation height, and that subtraction underpins any risk analysis.
- Orthophoto: the RGB imagery corrected and draped on the DTM, at scales that in our case run from 1:500 to 1:50,000 depending on what needs to be seen.
- 3D corridor model: towers, conductors modelled as catenaries, insulator strings, crossings with roads and other lines.
- Inventory: every tower with its position, type, height and visual condition; every span with its length and sag.
- Clearance analysis: the minimum distance of each span to vegetation, ground, buildings and other infrastructure, checked against the regulations.
The last one pays for the project. The Spanish high-voltage line regulation (Royal Decree 223/2008, ITC-LAT 07) sets the clearance to trees at 1.5 metres plus the electrical distance Del, with a minimum of 2 metres. From the classified cloud that distance is computed for every tree and every conductor at maximum sag, and out comes a list of spans ranked from worst to best. An 80-kilometre line may have 300 spans and, in a normal year, a handful in breach and a few dozen about to be.
Day 12: the decision
Here our part ends and the client's begins. The report does not say “you have a vegetation problem”; it says which spans, which trees, how many metres and in what order. With that, the operator contracts pruning where it is needed rather than where it has always been done, brings forward the inspection of a splice the thermal camera flagged three degrees above its neighbours and, if the orthophoto shows someone has put up a warehouse under the line, opens another file. They also decide what not to do, which is the cheapest decision and the hardest to justify without data.
At Azimutal we have captured more than 10,000 kilometres of power lines with our own sensors, calibrated and synchronised with each other, and classified them with our own software. We deliver the classified cloud, the DTM, the orthophoto, the 3D corridor model, the tower inventory and the clearance analysis, in whatever format the client's system understands. What we do not deliver is a pretty flight with no conclusions.
A point cloud decides nothing; the decision belongs to the person who finally knows exactly where every tree stands.
Sources
- Proyecto PNOA-LiDAR: coverages, densities and accuracies — IGN / MITECO, 2022
- Technical specifications for LiDAR flights, 3rd PNOA-LiDAR coverage — IGN, 2023
- Estado actual y futuras actuaciones del programa PNOA-LiDAR — IGN, 2018
- RIEGL VQ-1560 II-S datasheet — RIEGL, 2020
- Applanix POS AV specifications — Trimble Applanix
- Big LiDAR Data in Digital Earth — CEUR Workshop Proceedings, 2020
- Lidar point classification (ASPRS LAS codes) — Esri ArcGIS Desktop documentation
- Electrical clearances under RD 223/2008 ITC-LAT 07 — FEMPA, 2016
- Real Decreto 223/2008, regulation on technical conditions for high-voltage power lines — BOE, 2008



