Advances in the technology of automatic mapping of forests

This image shows the input and output data of the tree segmentation algorithm. The input data (left) are colored by height. The algorithm results (right) use color to segment each tree from the point cloud. (Purdue University, Joshua Carpenter)

WEST LAFAYETTE, Ind. – How lightning travels from the sky to the ground inspired the concept behind a new algorithmic approach to digitally separate individual trees from their forests in automatic forest mapping.

“As lightning travels from heaven to earth, it finds the path of least resistance through the atmosphere,” said Joshua Carpenter, a graduate student at Lyles School of Civil Engineering in Purdue. This got him thinking the same way about his digital forest data or point clouds.

“If I could somehow treat all the points in this point cloud like a path of least resistance, that would tell me something about where the tree is,” Carpenter said. The concept also works from a plant-biological point of view.

“Every leaf in a tree needs to be fed nutrients, and the nutrients come from the soil. This is how we find the shortest route for tree nutrients from the canopy to the ground.”

Carpenter and four Purdue co-authors recently published the details of their mapping methods in the journal Remote Sensing. The approach means the difference between mapping a few trees and mapping hundreds of acres at once quickly and with great accuracy. It could also lead to creating digital twins of forests, which could improve management planning in the face of climate change, disease outbreaks and population growth.

The work was supported in part by Purdue’s Integrated Digital Forestry Initiative. This initiative, one of five strategic investments in Purdues Next Moves, uses digital technology and multidisciplinary expertise to measure, monitor and manage urban and rural forests to maximize social, economic and environmental benefits.

“We have developed a new individual tree segmentation algorithm that can be used to perform a tree inventory for large areas,” said article co-author Jinha Jung, assistant professor of civil engineering. Carpenter is a member of Jung’s Geospatial Data Science Laboratory, which specializes in mapping and surveying.

“Another contribution of this paper is the evaluation of the performance of the segmentation algorithm with data collected from the ground,” said Jung.

The algorithm has proven to be highly accurate, often by a large margin, compared to the current state of the art for most metrics. The validation involves directly tagging and measuring individual trees in the field to correlate them with LiDAR data collected on the ground and from the air at different times of the year to capture leafy and leafless trees.

The team is still grappling with issues arising from their three data collection methods: photogrammetry (creating 3D images from 2D photographs) and two types of LiDAR (airborne and ground-based).

Data in the point cloud has the same structure, but the data from each method contains different anomalies. You can capture the details of the treetops quite well, but miss elements of the trunk and vice versa. Also sometimes included in landscape block data collection.

“The goal is to use all available point clouds to create a flexible algorithm,” explained Carpenter. “But finding a method to work with each of the specific anomalies is a challenge.”

The Purdue team operates in the 400-acre Martell Forest about 8 miles east of campus and continues to expand the scope of its technology.

“How can we go from several hundred acres to several thousand or several hundred thousand and then every tree on the planet? This is the future,” said article co-author Songlin Fei, professor and dean of the Chair of Remote Sensing in Forestry and Natural Resources. “The question is how to scale it.”

The inventory requires lengthy fieldwork to sample 5% or 10% of an area. “A 100 percent inventory has never been an option. This paper demonstrates technologies that allow counting of each individual tree. We’re talking about a huge leap,” Fei said.

The remote sensing paper focuses on mapping forests, but more algorithms are needed to create full inventories.

“With this data we can carry out diameter measurements. But what about other important stock characteristics such as straightness, wood quality or species identification? These have yet to be achieved,” said Fei.

The technologies now make it possible to create a digital twin of an entire forest to see the possible effects of an ice storm or high winds.

“When you make a forest management plan, you can’t just harvest the trees and see what it looks like,” Fei noted. “But in the digital world, you can cut down any tree you want and you can put it back. This enables simulations and better management planning.”

In the last decades geospatial data have enormously increased agricultural production. Purdue researchers are aiming to do similar things for forestry, a source of important raw materials for construction and fuel. Catastrophic wildfires and invasive species that have wiped out large stands of American chestnut and ash trees are now drawing attention to the importance of forests.

“We’ve successfully applied all of these technologies to agriculture,” Carpenter said. “But other areas need our attention now.”

– Purdue University Agricultural News

source

Leave a Reply

Your email address will not be published. Required fields are marked *