UAV LiDAR for Forest Inventory: Canopy Height, Biomass and Terrain Mapping

Forest inventory has long depended on manual ground plots—measurements that are time-consuming, costly, and inherently limited in spatial coverage. A typical field crew can sample only a tiny fraction of a management unit, leaving vast areas uncharacterized. UAV LiDAR changes this calculus by delivering direct 3D measurements of canopy structure and the ground surface beneath it, enabling accurate, scalable inventory across entire stands.

LiDAR-equipped drones fire laser pulses that penetrate dense canopies, capturing returns from foliage, branches, and the forest floor. The resulting point cloud supports precise canopy height models, aboveground biomass estimation, and digital terrain models—all from a single flight. Research in China’s subtropical forests demonstrated that UAV LiDAR enabled rapid, accurate biomass mapping across 1.72 km² using a DJI M300 RTK platform, achieving an average point density of 50.5 points/m² per this MDPI study.

Forestry agencies and private companies are increasingly adopting this technology for operational inventory, carbon accounting, and harvest planning. This guide covers the underlying principles, key applications, equipment selection, data acquisition workflow, and processing steps needed to deploy UAV LiDAR effectively. For a broader comparison of mapping technologies, see our guide on LiDAR vs. photogrammetry for drone mapping.

LiDAR—Light Detection and Ranging—measures distance by firing laser pulses toward the ground and timing their return. Each pulse that strikes a surface generates a 3D point, and millions of these points form a point cloud that captures forest structure in three dimensions. The critical advantage over passive sensors is that LiDAR pulses penetrate canopy gaps, producing returns from foliage, branches, and the forest floor in a single pass.

A UAV LiDAR system integrates three core components: the laser scanner, a GNSS/IMU unit for precise positioning and orientation, and the drone platform itself. The GNSS/IMU is essential—it georeferences each return so that the point cloud can be accurately positioned in real-world coordinates. Without it, the data would be unusable for inventory purposes.

Point density and scan angle directly influence canopy penetration and ground detection. Forest surveys typically operate at 10–50 points/m², with higher densities improving individual tree detection and terrain mapping. Dense canopies and leaf-on conditions reduce ground return rates, so flight planning must account for seasonal foliage. For a practical comparison of mapping technologies, see our guide on LiDAR vs. photogrammetry for drone mapping.

The canopy height model is the foundational product for most forest inventory metrics. To generate a CHM, the point cloud is first classified into ground and non-ground returns. The ground points are interpolated into a digital terrain model (DTM), and every remaining point is normalized by subtracting the terrain elevation beneath it. The result is a surface where each point’s value represents height above ground—the CHM.

Individual tree detection then applies local maxima filtering to the CHM, identifying treetops as the highest points within a moving window. Crown boundaries are delineated using marker-controlled watershed segmentation, which treats each treetop as a catchment basin per this study on tree crown delineation. In a conifer plantation with regular spacing, this approach reliably extracts individual tree heights, supporting stand-level volume estimates and growth monitoring.

Aboveground biomass (AGB) is estimated by relating LiDAR-derived metrics—tree height, crown area, and canopy density—to field-measured biomass through regression models. A study in China’s subtropical forests used a DJI M300 RTK with an integrated LiDAR system, collecting data at 50.5 points/m² across 1.72 km². The researchers extracted individual tree heights and diameters from the point cloud, then applied a binary empirical model combining both variables to map AGB at regional scale per this MDPI study. Reported accuracies typically reach R² values above 0.85 for height and 0.80 for biomass when validated against field plots, though performance degrades in multi-layered canopies where understory returns complicate height extraction.

LiDAR’s ability to penetrate canopy gaps makes it uniquely suited for terrain mapping in closed forests. Ground classification algorithms identify returns from the forest floor, which are interpolated into a continuous DTM. This capability is critical in rugged, densely vegetated terrain where ground surveys are impractical and photogrammetry fails due to vegetation occlusion. Point density requirements vary by objective: terrain mapping can succeed at 10–20 points/m², while individual tree detection in complex stands benefits from densities above 30 points/m². Leaf-off conditions improve ground return rates significantly, so winter flights are often preferred in deciduous forests.

Additional derived products include canopy fuel metrics—canopy cover, height, and bulk density—which support fire behavior modeling per this USDA Forest Service technical report, and carbon stock assessment for climate reporting. For operational guidance on sensor selection, see our guide to drone LiDAR payload suppliers.

Field plots remain the gold standard for ground-truthing, but they sample only a fraction of a management unit. A crew measuring 0.1-hectare plots might cover 1–2 hectares per day in rugged terrain, while a single UAV LiDAR flight can capture 50–100 hectares with consistent point density across the entire area. The technology also eliminates access constraints—steep slopes, dense understory, and remote stands that require hours of hiking become surveyable from the air. In China’s subtropical forests, researchers mapped 1.72 km² of complex terrain in four flights, extracting individual tree heights and diameters that met field-survey accuracy standards per this MDPI study. The trade-off: field plots are still required to calibrate biomass models and validate LiDAR-derived metrics, so UAV LiDAR complements rather than fully replaces ground work.

Photogrammetry derives 3D structure from overlapping RGB images, but it reconstructs only visible surfaces. Under dense canopy, the ground is occluded, producing a canopy-top surface model with no terrain beneath—a critical failure for terrain mapping and height normalization. LiDAR pulses penetrate canopy gaps to reach the forest floor, yielding both terrain and canopy returns in a single pass. Photogrammetry also depends on consistent lighting and texture; shadowed or homogeneous surfaces degrade matching quality. LiDAR is active, emitting its own pulses, so it operates regardless of illumination. For applications requiring accurate ground elevation under canopy, LiDAR is the only practical UAV-based option.

Satellite LiDAR missions like GEDI and ICESat-2 provide global coverage but sparse sampling—footprints of 25 m and 100 m respectively, with no wall-to-wall mapping. UAV LiDAR delivers centimeter-level point clouds at 10–50 points/m², resolving individual trees. Manned airborne LiDAR covers far larger areas per flight—thousands of hectares—but requires significant mobilization costs and is uneconomical for small-to-medium holdings. UAV systems offer comparable accuracy at a fraction of the cost for areas under a few hundred hectares, with the flexibility to re-fly on demand.

The main UAV constraints are battery endurance, weather sensitivity, and regulatory limits on beyond-visual-line-of-sight operations. For operational guidance on sensor selection, see our guide to drone LiDAR payload suppliers.

Sensor choice determines the quality and cost of your forest inventory. Four specifications matter most: point density (returns per square meter), range (maximum operating altitude), accuracy (vertical and horizontal error), and weight (which constrains drone selection). For forest work, prioritize sensors with multiple returns per pulse—these capture canopy layers and ground simultaneously. Scan angle also matters: wider angles improve coverage but reduce canopy penetration at oblique incidence.

The drone must lift the sensor while maintaining adequate flight time. A typical LiDAR payload weighs 1.5–3.5 kg, requiring a platform with at least 5 kg payload capacity for reasonable endurance. The GNSS/IMU quality is equally critical—it georeferences each return, and lower-grade units introduce horizontal drift that degrades tree positioning. Look for drones with RTK/PPK capabilities and replaceable batteries for extended missions.

The DJI L2 suits operators wanting an integrated solution with existing DJI drones, while the miniVUX-3 offers superior accuracy for individual tree detection on small drones per Balko Tech’s forestry applications guide. The VUX-120 excels at large-area terrain mapping where range and precision justify its higher cost.

Complete systems range from roughly $20,000 (L2 with drone) to over $150,000 (VUX-120 with survey-grade drone). Factor in processing software—commercial packages add $3,000–10,000 annually. For organizations starting out, the L2 offers the lowest entry barrier; for operational programs covering hundreds of hectares, the miniVUX-3 provides the best accuracy-to-cost ratio. For a broader market overview, see our guide to drone LiDAR payload suppliers.

Flight altitude, speed, and scan angle directly determine point density and canopy penetration. Lower altitudes increase point density but reduce coverage per flight; higher altitudes do the reverse. For forest surveys, typical parameters are 100–400 m above ground level with a scan angle of 30°, yielding 10–50 points/m² depending on sensor settings. A 100-hectare block flown at 120 m with a 50 m/s airspeed and 50% side overlap typically requires 3–5 flights on a multi-rotor platform, with each battery supporting 15–25 minutes of collection. The DJI M300 RTK used in China’s subtropical forest study flew at 300–400 m with a 30° scan angle, completing 1.72 km² in four flights at 50.5 points/m² per this MDPI study.

Foliage condition is the single largest factor affecting ground return rates. Leaf-off conditions in deciduous forests can double ground detection, improving DTM accuracy and height normalization. However, leaf-on data is required for canopy metrics like crown volume and leaf area. Where both terrain and canopy products are needed, plan leaf-off flights for DTM generation and leaf-on flights for canopy analysis—or accept reduced ground returns in dense stands.

LiDAR point clouds are georeferenced by the GNSS/IMU, so base station corrections or PPK processing are essential. In mountainous terrain, GNSS multipath from steep slopes degrades accuracy; placing ground control points at varying elevations helps validate vertical error. Weather constraints—wind above 10 m/s, precipitation, and low cloud—can abort missions, so build buffer days into the schedule. For operational guidance on accuracy requirements, see our UAV mapping accuracy guide.

Raw LiDAR point clouds contain returns from canopy layers, understory, and the ground surface. The first processing step is classifying each return as ground or non-ground using algorithms like progressive morphological filtering or cloth simulation. Ground points are interpolated into a digital terrain model (DTM), then every remaining point is normalized by subtracting the terrain elevation beneath it—converting absolute elevations into heights above ground. This normalization is essential because all subsequent metrics, from canopy height to biomass predictors, assume a common vertical reference.

The normalized point cloud is rasterized into a canopy height model (CHM) by taking the maximum height within each grid cell, typically at 0.5–1 m resolution. Voids caused by low return density are filled through interpolation. Individual tree detection applies local maxima filtering to identify treetops, then marker-controlled watershed segmentation delineates crown boundaries—treating each treetop as a catchment basin in an inverted CHM per this study on tree crown delineation. The output is a tree list with heights and crown areas that feeds directly into volume and biomass calculations.

Aboveground biomass (AGB) is estimated by relating LiDAR-derived metrics to field-measured biomass through regression. Common predictors include:

• Height percentiles (p50, p90, p95)
• Maximum and mean canopy height
• Canopy cover and density
• Crown area and volume

In China’s subtropical forests, researchers extracted individual tree heights and diameters from UAV LiDAR point clouds, then applied a binary empirical model combining both variables to map AGB across 1.72 km² per this MDPI study. Reported accuracies typically reach R² values above 0.85 for height and 0.80 for biomass when validated against field plots, though performance degrades in multi-layered canopies where understory returns complicate height extraction.

Open-source tools handle most forest LiDAR workflows: LAStools and PDAL for classification and normalization, R packages (lidR, ForestTools) for CHM generation and tree segmentation, and Python for custom modeling. Commercial options like TerraScan and FUSION offer streamlined interfaces and production-grade classification. The choice depends on team expertise and processing volume—lidR, for example, supports the entire pipeline from point cloud to tree list in a single scripted workflow.

Validation against field plots is mandatory. Compare LiDAR-derived tree heights and biomass estimates with ground measurements, reporting root mean square error (RMSE) and bias. Common error sources include low ground return rates in dense canopies, which inflate DTM elevation and underestimate tree heights; GNSS drift in steep terrain; and segmentation errors in overlapping crowns. Leaf-off flights improve ground detection and reduce these errors, particularly in deciduous stands. For operational accuracy requirements, see our UAV mapping accuracy guide.

A 2023 study in China’s subtropical forests demonstrated UAV LiDAR’s capacity for regional-scale biomass mapping. Using a DJI M300 RTK equipped with an integrated LiDAR system, researchers collected data across 1.72 km² in four flights at 300–400 m altitude with a 30° scan angle, achieving an average point density of 50.5 points/m². Individual tree heights and diameters extracted from the point cloud fed a binary empirical biomass model, enabling wall-to-wall aboveground biomass mapping. The extracted structural attributes met field-survey accuracy standards, confirming that UAV LiDAR can replace extensive ground measurement for regional inventory per this MDPI study.

A 2025 study across 14 one-hectare stands with varying tree densities combined UAV LiDAR with RGB imagery and deep learning. The unsupervised LiDAR pipeline applied marker-controlled watershed segmentation for crown delineation and local maxima filtering for tree heights, while a self-supervised RGB model learned crown boundaries from LiDAR-derived annotations—eliminating the need for extensive ground truth. This dual approach improved crown delineation accuracy in variable-retention harvest stands where canopy complexity degrades pure LiDAR segmentation per this study on direct AGB estimation.

USDA Forest Service research at Capitol State Forest used LiDAR-derived metrics—canopy cover, height, crown bulk density, and canopy base height—to generate 30-m resolution canopy fuel maps supporting fire behavior modeling and fuel mitigation programs per this USDA Forest Service technical report.

These cases share a common lesson: UAV LiDAR delivers inventory-grade accuracy at a fraction of the field effort, but success depends on matching point density and segmentation strategy to stand complexity.

The most significant limitation of UAV LiDAR is reduced ground return rates in dense, multi-layered canopies. In closed tropical or old-growth stands, fewer than 10% of pulses may reach the forest floor, degrading DTM accuracy and inflating tree height estimates. Leaf-on conditions in deciduous forests compound this problem. Mitigation strategies include scheduling leaf-off flights for terrain mapping, increasing point density to 30–50 points/m², and using sensors with multiple returns per pulse. In China’s subtropical forest study, a 30° scan angle and 50.5 points/m² density achieved sufficient ground penetration for accurate terrain modeling per this MDPI study.

Battery endurance limits coverage to roughly 50–100 hectares per day, and flights are weather-dependent—winds above 10 m/s, rain, or fog abort missions. Regulatory restrictions on beyond-visual-line-of-sight operations further constrain large-area surveys. Data processing demands specialized skills in point cloud classification and statistical modeling, which many forestry teams lack in-house.

• Validate GNSS/IMU calibration before each mission
• Place ground control points across varying elevations
• Collect field plots for biomass model calibration
• Assess ground return rates before accepting DTM outputs
• Document error sources and report RMSE against field data

Following these practices ensures UAV LiDAR delivers inventory-grade accuracy despite its operational constraints.

Can UAV LiDAR penetrate dense canopy to map the ground?

Yes, but ground return rates vary with canopy density and foliage condition. In closed, multi-layered stands, fewer than 10% of pulses may reach the forest floor. Leaf-off flights, higher point densities (30–50 points/m²), and sensors with multiple returns per pulse improve terrain detection.

What accuracy can I expect for canopy height and biomass?

Reported height RMSE typically ranges from 0.5–1 m when validated against field plots. Biomass models using LiDAR-derived height and diameter metrics commonly achieve R² values above 0.80, though accuracy degrades in complex, multi-layered canopies.

How does UAV LiDAR compare to photogrammetry for forest mapping?

Photogrammetry reconstructs only visible surfaces and fails to map terrain under dense canopy. LiDAR pulses penetrate gaps to capture both canopy and ground returns in a single pass, making it the only practical UAV-based option for terrain mapping in closed forests.

What equipment is best for forest inventory?

The DJI Zenmuse L2 suits small stands with an integrated workflow; the RIEGL miniVUX-3 offers superior accuracy for individual tree detection on small drones. For large-area terrain mapping, the RIEGL VUX-120 provides the range and precision needed.

How long does data processing take?

Processing time depends on area and software. A 100-hectare block typically requires 1–3 days for classification, normalization, and tree segmentation using tools like LAStools or the R package lidR.

What regulations apply to forest drone flights?

Beyond-visual-line-of-sight operations are restricted in most jurisdictions, limiting coverage per mission. Check local aviation authority rules and obtain necessary waivers before planning large-area surveys.