How accurate is LiDAR forest inventory in published trials?

At the stand level, about as accurate as a field cruise, when both are checked against what was actually harvested. A 2025 trial in a mixed conifer forest checked a cruise and airborne laser scanning inventories against logs scaled at the mill. The laser methods produced "stand-level volume estimates with similar errors". They ranged from minus 8 to 6%, against minus 16 to 6% for the cruise. Below the stand, accuracy falls away. The same group found laser methods missing trees in a single stand, and a 2023 validation found volume by diameter class often wrong by more than 100%.

What the usual answer says

The usual answer quotes LiDAR accuracy as agreement with field measurements of height, diameter and volume, often within 5 to 15% for stand volume. Those comparisons are real. They leave out two things: which level of question the figure answers, and how accurate the field reference was.

Stand totals against mill scale

Sparks, Corrao, Keefe, Armstrong and Smith, in Forests, start from a gap. They write that "few studies have quantified errors in field sampling- and airborne laser scanning (ALS)-derived inventories at the stand level". They built four inventories of the same forest: a cruise of variable-radius plots, an area-based laser model, and two methods that find individual trees. Each was checked against "the gross volume of harvested logs from multi-stand harvest data, tracked by load and location and scaled at the processing mill".

The laser and cruise errors were similar. Across the forest, the cruise agreed more closely with the individual tree methods than with the area-based model. The authors suggest both may undercount trees in parts of it. They also note what laser inventories add: "the spatial variability of within-stand attributes that ALS inventories provide". We read the abstract only, because the publisher's site could not be opened, so stand counts and per-stand errors are not given here.

Individual trees against felled trees

The same authors, in a 2024 Forest Science study, went further in a single 1.1 ha stand. They measured felled trees directly and followed the logs to the mill. Laser heights had lower error and bias than field heights. But the laser methods "underdetected trees". They accounted for 78% to 91% of the merchantable volume in the field reference, and 71% to 99% of "the merchantable volume scaled at the mill". So at tree level, laser inventory missed as much as 29% of what was scaled. The authors add that the study illustrates "challenges of using mill-scaled volume estimates as validation data". We read this one as an abstract too.

Volume by diameter class

Strunk and McGaughey, in a 2023 stand validation of an area-based model for a managed pine forest, found that "Stand-level results were consistently better than pixel-level results", by 10 to over 200 percentage points. Totals held up. Product detail did not. Volume predictions for specific diameter classes "often fared poorly", with errors above 100% RMSE, "especially for larger (less common) diameter trees". The authors note that in practice diameter data are rarely used one narrow class at a time, and speculated that wider classes might be acceptable.

How good is the reference?

Most accuracy figures, in our reading of the literature, compare laser estimates with field plots. Those plots have errors of their own. Noordermeer, Gobakken and colleagues, in Silva Fennica (June 2025), used 12,420 plots of 250 square metres from a national inventory and 45 local management inventories. They varied how sample trees were chosen and how plot values were calculated, for timber volume, Lorey's mean height and dominant height. Root mean square errors "ranged from 5% to 16% relative to the mean observed values, across the factors studied". The best accuracy came from choosing sample trees in proportion to basal area and keeping their measured heights. Plot data are, in the study's words, "an essential reference for calibrating predictive models", so plot error feeds into the laser model as well as into judging it.

Reading it through a data lens

An accuracy figure means little without its reference and its level. Within 8% of mill scale for a stand and within 8% of field plots for a pixel are different claims. Neither says anything about volume in the 40 cm class. The fairest reference an operation holds is its own harvest record, where scale tickets can be tied to a stand and a load. The 2025 trial was possible because harvest data had been "tracked by load and location". An operation that keeps that link can run the same test on any inventory it buys.

When it doesn't apply

Where harvest volumes can't be tied to stands, the mill-scale test isn't available, and field plots remain the practical reference. In our judgement, scaled volume measures merchantable removals, not standing volume, since breakage and cull left in the woods never reach the mill. And trials in one forest type don't settle accuracy for others.

Quarri for forest management is built around how a forest operation runs, from the cruise to the settled account.

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.