📄 other

Using UAV-LiDAR for stem volume phenotyping in the genetic selection of radiata pine

This study demonstrates that UAV-LiDAR-derived metrics, particularly 3D convex hull surface area and tree height, serve as highly heritable and scalable proxies for stem volume that can effectively accelerate genetic gain in radiata pine breeding programs while reducing reliance on labor-intensive field measurements.

Original authors: Joane Simone Elleouet, Sadeepa Jayathunga, Natalie G Graham, Mark F Paget, Jaroslav Klápště, Robin J.L. Hartley

Published 2026-07-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Joane Simone Elleouet, Sadeepa Jayathunga, Natalie G Graham, Mark F Paget, Jaroslav Klápště, Robin J.L. Hartley

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to find the "superheroes" of the forest. In the world of tree breeding, scientists are constantly looking for the specific trees that grow the biggest, straightest, and fastest, so they can be used to plant forests that produce more timber. But here's the tricky part: trees take decades to grow up. If you wait until a tree is fully grown to see if it's a winner, you've wasted a whole generation of time. So, breeders have to guess which young saplings will become giants based on clues they can see right now.

Traditionally, the main clue has been the tree's "waist size," measured at chest height. This is called Diameter at Breast Height, or DBH. It's like measuring a person's waist to guess how much muscle they'll have when they're an adult. But measuring thousands of trees with a tape measure is slow, boring, and expensive. Plus, you can't measure the whole tree's volume just by looking at its waist; you need to know how tall and bushy it is, too.

Enter the new technology: drones equipped with laser scanners (LiDAR). Think of these drones as flying robots that shoot millions of invisible laser beams at the forest. These beams bounce back to create a super-detailed 3D map of every single tree, capturing its height, shape, and how much space it takes up in the air. The big question for scientists has been: Can these laser maps replace the old-fashioned tape measures? Can a drone spot the future giants just as well as a human with a tape measure, but much faster?

This is exactly what a team of researchers set out to find in a recent study on Radiata pine, a fast-growing tree famous in New Zealand. They didn't just guess; they flew drones over 11 different forest trials containing about 27,000 trees. They compared the drone's laser measurements against the "gold standard" of tree volume, which is calculated using the tree's waist size (DBH) and its height.

The researchers discovered that the drone is incredibly good at spotting the trees with the most "woody bulk." While the drone's measurement of a tree's total height was the most reliable single number to predict how fast a tree would grow, the real star of the show was a specific 3D measurement called the "convex hull surface area." Imagine wrapping a tree in a tight, invisible plastic bag that hugs every branch and leaf; the surface area of that bag is the convex hull. The study found that this 3D "bag size" was the best laser-based guess for how much wood a tree would eventually have, matching the traditional measurements with a very strong connection (a correlation of up to 0.87).

However, the paper also rules out the idea that drones should completely replace humans just yet. The researchers found that while the drone is amazing, it has a blind spot: it can't easily tell if a tree is "broken" or "malformed" (like having two tops instead of one) just by looking at the laser map. In the forest, a broken tree is a loser, but the drone might still think it's a giant because it's big and bushy. When the scientists included these broken trees in their drone-only analysis, the accuracy dropped a little bit.

So, what's the verdict? The paper suggests that the best strategy isn't to choose between the drone and the tape measure, but to use them as a dynamic duo. The drone can quickly scan thousands of trees to find the ones with the best height and 3D shape, acting as a super-fast filter. Then, breeders can use the traditional tape measure on the top candidates to double-check for defects and get the final waist measurement. This combination allows them to spot the genetic superstars much earlier and cheaper than before, potentially speeding up the creation of better forests without losing the accuracy needed to make the right choices. The drone doesn't replace the detective; it just gives the detective a high-tech magnifying glass to solve the case faster.

Technical Summary: Using UAV-LiDAR for Stem Volume Phenotyping in the Genetic Selection of Radiata Pine

Problem Statement
Radiata pine (Pinus radiata) breeding programs prioritize stem volume at harvest age as a primary objective. However, the long rotation periods (25–30 years) in New Zealand necessitate indirect selection based on juvenile traits to accelerate genetic gain. Traditional phenotyping relies heavily on Diameter at Breast Height (DBH) as a proxy for stem volume, often combined with field-measured height in allometric equations. This approach faces significant constraints: field measurements are labor-intensive, costly, and time-consuming, creating a phenotypic bottleneck that limits selection intensity. Furthermore, ground-based height measurements are often imprecise compared to DBH, and allometric equations can introduce relative errors up to 20% due to tree-to-tree variation in stem form. There is a need for scalable, high-throughput alternatives that can accurately estimate stem volume and capture genetic variation not fully reflected by traditional diameter measurements.

Methodology
The study evaluated the utility of Unmanned Aerial Vehicle (UAV) LiDAR as a phenotyping tool across 11 clonally replicated full-sib genetic trials of radiata pine in New Zealand, encompassing approximately 27,000 trees from two trial series (BC52-55-59 and BC66).

Data Acquisition: UAV-LiDAR data were collected using a DJI L1 sensor mounted on a DJI M300 RTK platform. Flight parameters were optimized to generate dense point clouds (45 m altitude, 3–5 m/s speed). Field data, including DBH and defect records, were collected concurrently or within three months of the UAV flights.
Metric Derivation: Individual tree metrics were extracted from segmented point clouds. Key metrics included LiDAR-derived tree height ( $z_{max}$ ) and geometric descriptors such as 2D and 3D convex hull surface areas and volumes.
Reference Trait: Allometric Stem Volume (ASV) was calculated using a validated equation (Kimberley and Beets, 2007) combining field-measured DBH and LiDAR-derived $z_{max}$ . This served as the reference for stem volume.
Genetic Analysis: A single-step genomic best linear unbiased prediction (ssGBLUP) model was employed using ~9,500 SNPs and pedigree data. The model estimated breeding values (EBVs) for DBH, ASV, and various LiDAR metrics.
Comparative Analysis: The study compared the performance of LiDAR metrics against DBH and ASV by assessing:
- Phenotypic and genetic correlations.
- Narrow-sense heritability ( $h^2$ ).
- Overlap in the top 100 selected genotypes.
- The recovery of potential genetic gain when selecting based on LiDAR proxies versus DBH.
- The impact of including malformed trees (which are typically filtered out in field-based ASV calculations) in LiDAR-only analyses.

Key Contributions and Results

Optimal LiDAR Proxy: Among the tested metrics, the 3D surface area of the individual tree convex hull (convexhull3D_area) demonstrated the highest correlation with ASV (up to $r = 0.87$ ). It outperformed LiDAR tree height ( $z_{max}$ ) as a proxy for volume, with an average correlation of 0.77 across sites.
Heritability: LiDAR tree height ( $z_{max}$ ) exhibited the highest heritability across all sites (mean $h^2 = 0.37$ , range 0.23–0.50), indicating strong genetic control and less environmental noise compared to diameter. The heritability of convexhull3D_area was similar to that of DBH (mean $h^2 = 0.26$ ).
Selection Efficiency:
- Selecting the top 100 genotypes based on DBH recovered 88–96% of the potential genetic gain for ASV.
- Selecting based on convexhull3D_area recovered 67–86% of the potential ASV genetic gain.
- Crucially, convexhull3D_area identified 24–33% of top-performing ASV genotypes that were not selected by DBH, suggesting LiDAR captures unique genetic variation related to tree form and volume.
Impact of Malformations: When analyses included malformed trees (a scenario relevant to fully automated LiDAR phenotyping where visual defect scoring is absent), the performance of LiDAR proxies as ASV indicators decreased marginally (by 3.5–10.3% in recovered gain), but they remained effective proxies.
Model Validation: The ssGBLUP model using a blended pedigree-genomic relationship matrix ( $H_{0.5}$ ) provided a better fit and higher additive genetic variance estimates compared to a pedigree-only (PBLUP) model, validating the use of genomic data in these trials.

Significance and Claims
The paper claims that UAV-LiDAR-derived metrics, particularly convexhull3D_area and tree height ( $z_{max}$ ), possess desirable properties to complement, rather than immediately replace, traditional field phenotyping in radiata pine breeding.

Complementarity: While DBH remains the most effective single criterion for stem volume selection due to its direct relationship with the allometric equation, LiDAR metrics offer high heritability (especially height) and the ability to detect genetic variation missed by diameter alone.
Scalability and Cost: The study highlights that LiDAR provides a scalable, repeatable, and non-destructive method for measuring growth traits. This can reduce phenotyping costs and enable more frequent measurements, facilitating the characterization of growth trajectories.
Accelerated Genetic Gain: By integrating high-heritability LiDAR traits into multi-trait genetic evaluations, breeding programs can improve the accuracy of breeding value predictions, particularly in trials where DBH heritability is low. This supports earlier selection decisions and accelerates the delivery of improved genotypes.
Current Limitations: The authors modestly note that LiDAR metrics are currently sensitive to canopy density (reducing accuracy in older, closed stands) and cannot yet identify stem malformations without visual inspection. Therefore, the optimal strategy identified is a hybrid approach where LiDAR complements field measurements to enhance selection accuracy and efficiency.

More like this