Using iPhone LiDAR scans for mixed forest mensuration | Faculty of Environmental and Urban Change

by Tarmo Remmel

A critical component of compiling forest stand inventories requires the determination of tree dimensions (height, diameter, volume) of single trees (both standing or fallen, alive or dead) along with their geographic locations, species, and other attributes. These metrics are often correlated with tree age and growth characteristics. Further, using allometric equations, these tree measurements can be used to estimate the biomass of leaves, stems, and roots. However, field measurement of these metrics is laborious and time consuming.

Figure 1. An unrectified 3D LiDAR point cloud (~179 million points) of a 400 m2 forest plot obtained by a hand-held, mobile device. Points have been “painted” with the corresponding colour and texture obtained from simultaneously collected digital images.

Our goal was to develop, document, and demonstrate a workflow for converting rapidly collected detailed 3D point cloud characterizations of forest plots (Figure 1) from hand-held, mobile LiDAR (Light Detection and Ranging) available on iPhone and iPad Pro models, into a georeferenced GIS data layer that permits maps of tree stem locations and their attributes to be constructed. We hypothesize that this would simplify the data acquisition process while producing accurate and consistent results.

Figure 2. Ganesa Persaud (a) was hired in the summer of 2025 to help develop and implement the workflow for georectifying LiDAR scans of our forest plots. Sven Huyke (b) worked as a field technician collecting LiDAR scans and will continue with this work in his M.Sc. research.

During the summer of 2025, we collected LiDAR scans for 22 forest plots measuring 400 m² each (20 m × 20 m) in predominantly mixed-wood forest stands in Southern Ontario. Scans were collected using the Modelar application and downloaded for processing with CloudCompare and QGIS software. Ganesa Persaud (Figure 2a), an undergraduate student, was hired to develop, test, and implement a workflow for the georectification of these point clouds using known GPS locations that we had measured for plot corners and three additional leaf-litter collection baskets within each plot.

Figure 3. A georectified LiDAR scan aligned with field-collected GPS data for plot corners and leaf-litter collection baskets. The view of the 3D point cloud shown is as if viewing the plot from the top down (nadir).

This work proved more challenging than we had initially anticipated, as simply transferring and handling the 4-8 GB files, opening them, panning around, and interacting with the data required substantial computing resources. We also realized that some scans were flipped relative to others, resulting from scanning devices being held in various orientations. Once the plot orientations were sorted out using known locations of features within each plot, the georectification process, although requiring tremendous attention to detail, was implemented.

Figure 4. A cross-sectional view over a single tree stem. The void is an area with no LiDAR point reflections due to this being the inside the tree. Identifying algorithms to separate the void from its surroundings will greatly facilitate the automated construction of forest inventory maps.

This required identifying precise locations on the 3D scans that corresponded with known GPS locations to setup the spatial transformation to a real-world coordinate system. Ganesa’s work resulted in our ability to bring LiDAR scans together into their correct spatial reference and to integrate that data with other spatially coincident data layers (Figure 3).

Work continues with Sven Huycke (Figure 2b) to process these georectified scans and to compare measurement results with those collected using traditional field methods. His current work is devising and automating a method for collapsing the 3D point cloud into a 2D plane, such that the plane represents a nominal 1.3 m height above the local terrain where the diameter at breast height (DBH) is normally measured. Then, using a kernel point density estimation to identify where there are voids in the point cloud (Figure 4), tree stem locations can be mapped along with the diameter, circumference, basal area, and cross-sectional shape of each tree.