LiDAR — Light Detection And Ranging — measures distance by firing a laser pulse at something and timing how long the reflection takes to come back. Do that a few hundred thousand times a second while sweeping the beam across a landscape, and record where the sensor was and which way it was pointing for every shot, and the result is apoint cloud: a few million to a few billion measured positions in three dimensions.
That is the whole idea. Everything else — returns, intensity, classification, the LAS format — is bookkeeping around it. What follows is the bookkeeping, because it is the part that decides whether the data is usable.
Measuring a distance with light
Light covers roughly 0.3 metres every nanosecond. A pulse sent to a target 150 m away and back has travelled 300 m and taken about a microsecond. The range is simply half the round trip:
range = c × t ÷ 2
The factor of two matters more than it looks. It means every timing error is halved in the range — but it also means the timing has to be extraordinarily good. One millimetre of range is 6.7 picoseconds of round-trip time. This is why the interesting engineering in a laser scanner is in the clock and the detector rather than the laser, and why two scanners with identical published ranges can differ substantially in how clean their data looks.
Not every sensor times a pulse directly. Many short-range terrestrial scanners use a continuous modulated beam and measure the phase shift of the returning signal instead, which gives excellent precision at high speed but only works out to a range where the phase is unambiguous — typically a few hundred metres at most. Long-range and airborne scanners use pulses.
From a range to a coordinate
A range and a beam direction give you a point relative to the sensor. To place that point on the Earth you also need to know, at the instant the pulse was fired, exactly where the sensor was and exactly how it was oriented. On an aircraft that comes from two instruments working together:
- A GNSS receiver for position, usually differentially corrected against a ground reference station, giving a trajectory good to a few centimetres.
- An inertial measurement unit for orientation — roll, pitch and heading — sampled hundreds of times a second, because an aircraft's attitude changes continuously and an angular error becomes a position error that grows with altitude.
This is worth dwelling on, because it explains most of the error budget in airborne LiDAR. At 1,000 m altitude, an orientation error of 0.01° puts a point about 17 cm out of place. The laser is not the limiting factor; the trajectory is. It also explains the artefacts you will eventually meet in real data: a constant angular offset between the scanner and the IMU (the boresight error) makes overlapping flight lines disagree, tilting in opposite directions, and has to be calibrated out before the strips are merged.
Terrestrial scanning sidesteps all of this. A scanner on a tripod does not move, so there is no trajectory to solve — but the scan then sits in its own local coordinate system, and the equivalent problem becomes registration: working out how each scan position relates to the others. That is the problem theE57 format is built around.
Why one pulse makes several points
A laser beam is not a line. It diverges as it travels, so a beam a few millimetres across at the aperture arrives as a footprint of perhaps 20–50 cm from typical survey altitude. That footprint can straddle several surfaces at different distances, and the sensor records a separate return for each echo it can resolve.
A pulse fired at a tree is the standard example. It clips an upper branch (return 1), passes through a gap to lower foliage (return 2), and finally reaches the ground through another gap (return 3, the last return). Three points, one pulse, from a single line of sight.
This is the property that makes LiDAR different from photogrammetry. A photograph only records the first thing the light hit. LiDAR sees past partial obstructions, which is why it can map terrain under a forest canopy, and why an archaeological survey can find earthworks under woodland that are invisible from the air. It is also why a photogrammetric point cloud from a drone has no useful return information at all — a limitation that shapes how such clouds have to beclassified.
There are limits. Two surfaces have to be separated by roughly a metre or two before their echoes can be told apart, because the pulse has a physical length in the air and the detector needs time to recover. Dense low scrub often returns a single merged echo somewhere between the top of the vegetation and the true ground — which is exactly the case that makes ground classification hard.
How many returns a file can express depends on its point format: the legacy LAS formats cap the return number at 5, while the LAS 1.4 formats allow up to 15. LAS point formats covers which is which.
What each point carries
A point is not just a coordinate. The LAS standard defines a set of per-point fields, and which ones are present depends on the point data record format the file uses.
| Field | What it is | Worth knowing |
|---|---|---|
| X, Y, Z | Position in the file's coordinate system | Stored as 32-bit integers with a scale and offset from the header, not as floating point. This is why the header matters. |
| Intensity | Strength of the returned signal | Uncalibrated and not comparable between files. See below. |
| Return number | Which echo of its pulse this point is | 1 is the first return. Zero is not valid and means the writer never filled the field. |
| Number of returns | How many echoes that pulse produced in total | A point where these two are equal is a last return. |
| Classification | What the point was judged to be | Ground, vegetation, building and so on — see the class table. |
| GPS time | When the pulse was fired | Lets points be traced back to a moment in the trajectory. |
| Scan angle | Beam angle from nadir | Points at the edge of a swath are less accurate; this is how you find them. |
| Point source ID | Which flight line or scan the point came from | Essential for diagnosing disagreement between overlapping strips. |
| RGB | Colour, where present | Almost never measured by the laser — usually draped from imagery afterwards. |
You can see all of these for a real file in thepoint cloud viewer, which colours by elevation, intensity, classification or RGB, and reports the distribution of elevation, intensity, returns and classes in itsstatistics panel.
A warning about intensity
Intensity is the amplitude of the returned pulse, and it is tempting to read it as reflectance — bright asphalt, dark water, and so on. Broadly that works. Precisely it does not, because the recorded value also depends on range, incidence angle, atmospheric attenuation, the scanner's automatic gain, and whatever scaling the vendor applied on export.
The practical consequence: intensity is useful within one flight line, misleading between flight lines, and meaningless between different surveys unless somebody has explicitly radiometrically calibrated it. Treat a visible intensity seam between two strips as an artefact of the sensor, not a change in the ground.
Where the data comes from
| Platform | Typical use | Character of the data |
|---|---|---|
| Airborne (crewed aircraft) | National mapping, flood modelling, forestry, corridors | Wide coverage, fairly even density, good ground penetration, little detail on vertical surfaces. |
| Drone / UAV | Site surveys, stockpiles, small corridors | Very high density over a small area. Lighter sensors, so the trajectory is usually the accuracy limit. |
| Terrestrial (tripod) | Buildings, heritage, as-built, tunnels | Extremely dense near the scanner, detailed facades, hard shadows behind every object, needs registration. |
| Mobile (vehicle-mounted) | Roads, rail, street furniture, city modelling | Dense along the route and nothing off it. Accuracy suffers wherever GNSS is obstructed. |
| Bathymetric | Shallow water, coasts, riverbeds | A green wavelength that penetrates water. Far lower density, and depth limited by water clarity. |
The distinction that matters most in practice is between airborne and terrestrial, because it changes what the data can tell you. An airborne cloud is effectively a very detailed surface seen from above. A terrestrial cloud is a set of interior and facade views with occlusion everywhere. Software, formats and classification schemes all lean toward one or the other, which is whyE57 and LAS both exist.
What it is used for
- Terrain models. Classify the ground returns, drop everything else, and interpolate — the route from a point cloud to aDTM or DSM, and the single most common reason LiDAR is commissioned.
- Flood and hydraulic modelling. Which depends entirely on bare-earth terrain, and therefore entirely on ground classification being right.
- Forestry. Canopy height is the difference between the first-return surface and the ground, both of which come from the same file.
- Corridor and asset surveys. Power lines, rail and roads, where the clearance between a wire and the vegetation under it is the deliverable.
- Construction and as-built verification. Comparing what was built against what was designed.
- Archaeology. Stripping vegetation to reveal earthworks that are invisible from the air and often invisible on the ground.
How the data reaches you
Almost always as LAS orLAZ. LAS is the ASPRS standard: a header describing the file, some variable length records carrying the coordinate system and other metadata, and then one fixed-size record per point. LAZ is the same content losslessly compressed to roughly a tenth of the size — LAS vs LAZ covers when to keep which.
COPC is a third arrangement: a LAZ file whose points are sorted into an octree so a viewer can fetch only the part it is looking at. Terrestrial work more often arrives asE57, which keeps scan positions separate rather than flattening them into one list.
Open a LAS or LAZ file in your browserto see all of this in one file, orinspect its header to read the coordinate system, point format and record layout without touching the points.
Common mistakes
- Assuming classification is correct. Classes are somebody's judgement, usually an algorithm's. A file that says class 2 is asserting ground, not proving it.
- Comparing intensity across files. As above. It is not a calibrated measurement unless someone made it one.
- Ignoring the coordinate system. A point cloud with missing or wrong CRS metadata still opens, still looks right on its own, and lands in the wrong part of the world the moment you combine it with anything else.
- Treating a first-return surface as terrain. It includes every roof and treetop. That is a DSM, not a DTM, and the difference is the subject ofits own article.
- Reading Z as height above sea level. Only if the file says so. The vertical datum is separate metadata from the horizontal one, and it is frequently absent.