Reading a multi-return LiDAR point cloud: what the returns actually tell you
First, intermediate and last returns are not just metadata. For a forestry or infrastructure survey they are the difference between a pretty picture and a bare-earth model you can design against.
Every laser pulse a survey-grade scanner fires can come back more than once. The first echo is whatever it hit first — a canopy top, a power line, a roof edge. If the beam keeps travelling through gaps in foliage it returns again from branches, understorey and, if you are lucky, the ground.
That sequence of echoes is what we mean by multi-return, and it is the single biggest reason we fly LiDAR instead of photogrammetry over vegetated terrain in Northland.
What each return is good for
| Return | Typical surface | Why it matters |
|---|---|---|
| First | Canopy top, structures, wires | Digital Surface Model, canopy height, clearance checks |
| Intermediate | Branches, shrubs, regenerating scrub | Vegetation structure, fuel loading, biomass proxies |
| Last | Ground (when penetration succeeds) | Bare-earth DTM, contours, cut/fill design |
| Single | Open ground, roads, water-edge | Hard surfaces where there is nothing to penetrate |
On a dense kānuka block you might see 20–30 % of pulses reach the ground. That sounds low until you realise photogrammetry sees none of it — the camera only ever models the top of the canopy.
Density is a design input, not a brag
We quote point density in points per square metre on the ground, after classification, not the raw figure the sensor logs. A 2016 regional dataset at 4–8 pts/m² is excellent for catchment-scale work. A skid-site or culvert design under canopy needs 100–400 pts/m², which is where a low, slow VTOL pass earns its keep.
If the deliverable is a DTM for design, ask the surveyor for ground-classified density. Everything else is marketing.
Classification: where the real work happens
Raw returns arrive unclassified. We run a ground filter tuned to the terrain type (steep gullies behave very differently to pasture), then manually check the edges — fence lines, stream banks, retaining walls — because that is where automated filters fail and where your contours will lie to you.
Standard ASPRS classes you will see in a Helix .LAZ:
2Ground3,4,5Low / medium / high vegetation6Building9Water14Wire conductor (where relevant)
Deliverables that respect the returns
- Classified
.LAZ— the source of truth. Keep it. - DTM / DEM as GeoTIFF at a cell size matched to ground density (not finer).
- Contours in
.DWG/.DXFand 12d-ready formats, generated from the DTM, not the raw cloud. - Canopy Height Model when forestry or clearance is in scope.
If you want to see what this looks like in a browser, spin the real Whangārei tile in our LiDAR data lab and toggle bare earth. The gullies that appear when the canopy drops away are the whole argument for multi-return.
