How We Scan a Block at Risk
Walking into a threatened stand with a SLAM sensor changes what you can know and how fast you can know it.
A block at risk has one defining constraint: we don't know how long it will still be standing. A hearing could be scheduled for October. A cutblock boundary could be revised next quarter. The felling contract has a tenure date. The work we do in those stands is shaped by urgency in a way that a baseline ecological survey is not.
That constraint pushed us toward SLAM. Simultaneous Localization and Mapping — the same family of algorithms that lets a self-driving vehicle build a map of its environment as it moves — has been adapted for mobile LiDAR platforms in forest environments. Paired with a camera sensor, our current rig can cover between 8 and 15 hectares per day in dense coastal rainforest. The equivalent tripod-based workflow, with stations 20–30 meters apart and 8-minute dwell times, covers closer to 2 hectares on a good day.
The Protocol
We enter the block from the lowest accessible point and work up-slope, running overlapping transects spaced 10–15 meters apart. The SLAM unit runs continuously, registering consecutive point clouds in real time using inertial measurement and feature matching. The camera captures a full-colour frame every 0.1 seconds, which fuses with the LiDAR geometry during post-processing to produce a coloured point cloud — geometry measured first, texture applied second.
At each transect end we run a short loop closure — retracing roughly 5 meters of a previously scanned path. Closing loops constrains drift in the inertial component. In dense canopy, GNSS is unreliable under the crown layer, so a separate GPS survey at open reference points on the block boundary provides the ground control needed for georeferencing.
After each day's session, the raw SLAM output is processed overnight on a laptop in camp. We check for drift artefacts, inspect the point density distribution, and flag areas needing fill passes. By end of week the result is a complete georeferenced point cloud of the block — typically 200–400 points per square meter at ground level.
What the SLAM Sensor Changes
The old workflow was instrument-led: you planted a scanner, scanned a hemisphere, moved, repeated. The forest shaped itself around your station positions. The SLAM workflow is body-led: you walk, and the instrument maps what you walk through. The forest shapes the path, not the equipment.
That shift has practical consequences. Dense cedar stands with no line of sight between stems — impossible to cover at station density without hundreds of setups — become walkable. Steep ravine walls that would require suspended equipment to scan from below become traceable from a safe traverse above. Understory complexity that would be occluded at tripod height becomes visible because the operator is moving through it at human height.
The fidelity is lower. Tripod LiDAR at a well-placed station produces sub-millimetre geometry on nearby surfaces. SLAM geometry at 200 points per square meter is sufficient for individual tree measurement — DBH, height, crown projection — but not for bark texture or fine branch structure. For the uses that matter most — stem inventories, basal area estimates, canopy structure analysis, legal filings — the resolution is more than adequate. For Gaussian splat visualization, we supplement with drone photogrammetry, using the SLAM point cloud as a georeferencing scaffold.
Prioritizing What We Scan First
When we enter a block with limited time, we do not scan a random sample. We work with the partner organization and, where applicable, the First Nation land office to identify which areas are most at stake — either because they carry the highest concentration of old-growth indicators, or because they sit closest to the active cutblock boundary.
TreeLearn runs on preliminary scans within hours of capture. The composition analysis tells us stem density, species distribution, and canopy height distribution in each scanned section. That information feeds back into the field schedule: where high old-growth indicator density appears, we return for denser coverage. Where a young fringe stand sits at a boundary, we document it and move on.
The output at the end of a two-week engagement is a mapped, measured, and analyzed block. Not an approximation — a permanent record.