What Actually Happens to a Point Cloud Before You See a Map

Most people meet LiDAR at the end: a clean contour map, a tidy building footprint layer, a surface that looks obviously correct. The work that makes it look obvious happens weeks earlier, and it is almost entirely unglamorous.

A raw scan is a few hundred million returns with no meaning attached. Every point knows where it is and how strongly it reflected. None of them know whether they hit tarmac, a roof, a power line or a bird.

Classification is the whole job

Ground filtering comes first, because nearly everything downstream depends on a correct bare-earth surface. Get it wrong and your contours inherit the error, your volumes are wrong, and your drainage analysis confidently routes water uphill.

The hard cases are predictable once you have seen a few: dense vegetation where almost nothing reaches the ground, steep slopes that filters mistake for buildings, and low walls that sit right at the threshold between ground and structure.

Noise is not random

Atmospheric returns, multipath off wet surfaces and the occasional flock of birds all produce points that are clearly wrong to a human and perfectly plausible to an algorithm. Removing them without also removing genuine low vegetation takes judgement.

Why this is worth outsourcing

A survey team that can fly a mission competently does not automatically want to own a processing pipeline, with the software licences, the workstation and the person who knows why a tile failed. Specialists like LiDAR Data Services exist because the processing is a different discipline from the acquisition, and the QA step in particular rewards having seen the same failure a hundred times.

Ask for the classified point cloud alongside the derived products. If a supplier will not hand it over, you cannot check their work, and you will not find out until something built from it does not fit.