People often think the hardest part of LiDAR is flying and firing the laser. In reality, the value is created on the ground, in the weeks after the survey — when, for each of billions of points, someone must decide whether it represents the terrain, a tree, a roof or a measurement error.
This transformation of the raw cloud into usable maps is called the geomatics processing pipeline. It explains why two LiDAR surveys of the same region can yield relief of very different quality: the laser supplies the raw material, the processing makes the product.
Step 1 — Georeference and assemble
The survey arrives cut into flight lines. First, each echo is combined with the GNSS and inertial trajectories to give it its final coordinates; then the overlaps between neighbouring lines are adjusted so they coincide to the centimetre. A poor alignment here and everything downstream inherits an offset.
Next comes the choice of reference system. In Québec, surveys are expressed in the MTM projection — the territory is divided into narrow zones that limit distortion. Each zone has its own code, and mixing sheets from different zones without reprojecting them is a classic mistake that warps distances.
Step 2 — Classify the points
This is the heart of the craft. Each point receives a class: ground, low, medium or high vegetation, building, water, bridge… Automated algorithms do the bulk of the work. One of the most common assumes that the ground is the lowest continuous surface: it gradually builds a mesh by adding points that could plausibly be terrain, and rejects those that “float” too high — foliage, wires, roofs.
To classify is to answer one question billions of times: does this point touch the ground, or something resting on it?
The quality of this classification decides everything. An overly aggressive algorithm shaves off real banks; too permissive, it lets shrubs contaminate the terrain. Difficult areas — cattail marshes, steep slopes, bridges — often require manual review.
Step 3 — Interpolate a continuous surface
Ground points, however dense, leave gaps. To obtain a regular grid — one elevation per one-metre cell — you interpolate: the height of empty cells is estimated from neighbouring points. The resampling must stay gentle; too coarse a choice creates staircase steps, invisible when flat but glaring in 3D.
Three products, three questions
From the same raw material come three complementary models:
- The DTM — digital terrain model. The bare ground, stripped of everything that covers it. This is the flagship product: hydrology, civil engineering and archaeology all rely on it. It is the one Québec 3D renders in relief.
- The DSM — digital surface model. The first return, that is, the top of everything: treetops, roofs, structures. Useful for solar exposure, line of sight, air corridors.
- The CHM — canopy height model. The difference between the two:
CHM = DSM − DTM. It gives vegetation height directly, the foundation of the modern forest inventory.
Resolution, precision, accuracy: don't confuse them
Three words wrongly used as synonyms. Resolution is the size of a cell (here 1 m). Precision measures the spread of repeated measurements. Accuracy says how closely you match the reality on the ground. A model can be finely resolved and wrong: display fineness never guarantees the correctness of the elevations.
Under the hood of Québec 3D
The relief you manipulate comes from the MRNF's 1 m DTMs, reprojected and encoded into elevation tiles the browser can read. No elevation is recomputed on the fly: you see the official data, given volume.
Open the map →The takeaway
Between the laser and the map, geomatics does invisible but decisive work: align, classify, interpolate. It is this pipeline that separates the ground from the rest and that, depending on its care, produces faithful or misleading relief. The next time you see a DTM as clean as a scan, remember that first, point by point, someone had to decide what was really the ground.