An example of the input and output data of the tree segmentation algorithm. At left, the input data is colored by elevation; at right, the results of the algorithm use color to segment each tree from the point cloud.

Going into detail: New algorithmic method improves tree mapping technology

Digital mapping applications continue to expand as the technology matures. Additional features and more granular visualizations make the technology valuable for planning and managing strategies for mitigating climate change. 

Purdue University engineers developed a new algorithmic approach to digital mapping that advances automatic forest mapping. The innovation could help to better understand how climate change-influenced disasters — strong storms or wildfires, for example — could impact forests and how to build resilience strategies.

Filling in the details

The study says that photogrammetry and two types of LiDAR, ground and aerial, are becoming the three go-to technologies for forest analyses and mapping. These are used to create a point cloud, a visual representation of points in space. 

The three-dimensionality of point clouds shows trees’ structural features, such as trunk diameter and canopy volume. This study introduces an automatic method for detailing individual trees in point clouds.

The virtual depictions of individual trees help the user better understand the trees’ orientation and characteristics. The program quickly and accurately maps acres of trees all at once, instead of just a few trees at a time. The features open up the possibility of creating a digital twin — a detailed virtual simulation of a real-world system — for an entire forest.

Four phases of the tree segmentation algorithm. In the first phase, the raw data is colored by height; the second phase shows individual tree stem data (trunks), and, in the final two phases, each tree trunk and its vegetation obtains a color.


The team compared their algorithmic results against tree data collected during fieldwork to test accuracy. They say the method is much more accurate by most metrics than other mapping technologies.

They’re still addressing some issues with their data collection technologies. Each one contains different anomalies, such as capturing canopy top details well but missing tree trunk characteristics.

“Coming up with a method to work with each of the specific anomalies is challenging,” Joshua Carpenter, a Ph.D. student, said in a news release. But the ultimate goal is to “use all of the different point clouds that are available to make a flexible algorithm.”

The team continues to perfect its innovation by working in a forest near Purdue and eventually wants to scale it up. Additional algorithms will be needed to achieve more complete digital tree inventories. For example, other algorithms could add data about tree species or timber grades.

Gathering tree inventory data generally requires tedious fieldwork to sample 5% to 10% of an area. This study demonstrates technology that could result in a complete census of all trees.

“A 100% inventory has never been an option. … We’re talking about a tremendous leap” with this innovation, said Songlin Fei, Purdue professor of forestry and natural resources. “How can we get from several hundred acres to several thousand or several hundred thousand, and then to every tree on the planet? That’s the future.”