A peer-reviewed study published on October 13, 2026 provides unusually practical evidence. Its researchers varied the flying and scanning conditions across 22 UAV LiDAR flights over a managed poplar plantation in southwestern France. Their best tree-height result had a 0.33 m root mean square error (RMSE) against terrestrial LiDAR. The best trunk-diameter result reached 3.2 cm RMSE. Those are best cases within a particular experiment, not general performance specifications for UAV LiDAR.
Why the visible canopy is the easier target
An overhead scanner has a comparatively clear view of the uppermost leaves and branches. Once enough points are collected, a canopy-height model can subtract a ground elevation model from the canopy surface. In the French experiment, height estimates became relatively stable around 1,000 points per square metre and at lower flight altitudes. Sparse 120 m flights tended to underestimate the canopy by roughly a metre relative to denser acquisitions.
DBH is a different reconstruction problem. The nominal measurement plane is 1.30 m above ground, beneath branches and often behind foliage, understory or neighboring stems. A scan can have thousands of points per square metre overall and still provide too few usable returns from a particular trunk cross-section. The study notes instances with fewer than 20 points in a stem slice at lower sampling densities. Fitting a circle or spline to those points can then be unstable or impossible.
The effect is visible in the error range. High-altitude, low-density acquisitions in the study produced trunk-diameter RMSE values of roughly 17.1 to 29 cm, while carefully selected lower flights were much closer to ground measurements. The authors also used a DJI Zenmuse L1 whose elongated beam footprint could bias the apparent width of stems. Better nominal positioning cannot restore geometry that the laser never observed.
Plan viewing angles, not just point counts
For regular plantation rows, the direction of the flight lines mattered. Paths aligned with the rows let more pulses reach stems through the gaps than equivalent cross-row paths. In a comparison of otherwise similar acquisitions, orientation produced a marked difference in diameter errors. This is a geometry result: the path determines which sections of each trunk are exposed to the sensor.
Overlap can also add useful angles. The strongest winter acquisition included 80% overlap, compared with 70% on otherwise related flights. It recovered all 33 measured trees and had dense stem slices. But the study tested the higher overlap in a limited set of flights; density and overlap were partly entangled, so 80% is not a universal prescription.
Survey margins are easy to overlook. A flight plan cut tightly to a plot boundary sees edge trees mainly from one side. The French team found that generous margins allowed a flight at 80 m with about 1,006 points per square metre to retain all measured stems and deliver better diameter estimates than a superficially similar survey with tighter coverage. The practical implication is to budget space for additional viewing angles at the perimeter, while respecting land access and operating constraints.
A separate 2025 study on stem detection and diameter also identifies flight altitude and LiDAR strip overlap as consequential. It reports that useful acquisition settings involve a balance among low flight height, speed and overlap. Neither paper establishes a universal recipe for every scanner, stand structure or species.
Returns help, but filtering changes what they buy
The Zenmuse L1 can record up to three returns per laser pulse. A later return may reach a trunk after an earlier interaction with the canopy. The French study observed that third returns contributed strongly to points below the upper crown, which helps explain their value for stem extraction.
There is a throughput trade-off: in the reported L1 settings, triple-return operation reduced pulse rate from 240 kHz to 160 kHz. Collecting more returns from each pulse does not necessarily mean more total points per area.
A subtler finding concerns processing. Before outlier removal, matched winter flights with three returns gave better diameter estimates than their dual-return counterparts. After filtering, the advantage in RMSE was much smaller. A procurement or mission-planning comparison that quotes only a sensor's maximum number of returns misses the effects of the processing pipeline. The proper comparison is the end-to-end result on the required metric, with the same validation set.
Leaf-off flights improve access to the stem
Timing changed the diameter results even when acquisition density was high. In the plantation study, the average diameter RMSE across three fitting methods rose from about 6.3 cm for a set of leaf-off flights with limited understory in winter 2024 to around 9.0 cm for spring 2025 flights with foliage and denser understory. The winter 2025 subset also performed worse than the previous winter as understory developed, which cautions against attributing the whole difference to leaves alone.
Canopy-height estimation was comparatively insensitive to the seasonal differences in the dense configurations tested. That finding should not be extended to sparse surveys or structurally different forests. It means that two forestry deliverables can have different optimal survey dates even when produced from the same instrument.
Another study of understory terrain estimation found stronger altitude sensitivity when canopy cover was at least 0.9. A height workflow still needs a trustworthy terrain model, particularly under dense crowns. The French poplar findings are therefore a useful case, not a license to ignore ground-return coverage.
The 1.30 m measurement is also a processing choice
The researchers compared three approaches to fitting trunk geometry: mean-circle fitting, maximum-inscribed-circle fitting and spline fitting. They examined short point-cloud slices between 0.60 and 2.00 m and selected the smallest fitted circumference as a proxy for a useful stem section. Comparison with field measurements found that this selection chose the closest circumference in 81% of cases, not in every tree.
That distinction matters. A standard DBH tape measurement is taken at a specific height, while the minimum-cross-section method searches a vertical interval. Taper, basal swelling, trunk shape and vegetation artifacts can move the selected slice away from the conventional measurement plane. Any deliverable labelled DBH needs its measurement convention documented, not hidden inside a point-cloud export.
The companion open dataset description records the 22 flights, settings, LAS products, field measurements and terrestrial LiDAR references. That enables a stronger test than judging point clouds from screenshots. A team can reproduce filtering and fitting decisions, then report stem detection success separately from error among successfully detected stems.
A testable survey specification
Before flying, define whether the client needs canopy height, stem diameter, terrain, or several of these at once. They have different acceptance criteria.
For canopy height, test the point density and ground model against known tree heights and check for systematic underestimation at the planned altitude. For diameter, inspect actual returns at approximately 1.30 m, flight-line orientation, plot-edge views, return mode, vegetation state and enough independently measured trunks to quantify both bias and RMSE. Report how many expected stems were not reconstructed; accuracy for only the easy trees can conceal an incomplete inventory.
Record sensor model, beam geometry, altitude relative to canopy and terrain, speed, scan mode, return mode, overlap, flight direction, boundary margin, season, processing software and filtering. Without those details, a quoted centimetre-level error is difficult to reproduce.
The October 2026 results came from 33 trees in one managed, regularly spaced 0.23 ha poplar plot, using an L1 and flight settings that were not a fully independent factorial experiment. Altitude, speed and density sometimes varied together. Mature mixed woodland, dense conifer stands and newer scanners need their own controlled checks.
The operational decision is straightforward: do not accept a convincing canopy model as validation of tree diameter. Design and score the acquisition for the hardest requested measurement, then confirm it with field references before turning the estimates into inventory or biomass figures.