Forest Inventory using high-density point clouds
Faculty mentor/Supervisor
Bogdan Strimbu
Email Address
Department Affiliation
Forest Engineering Resources & Management
Project Location
McDonald-Dunn research Forest
Project Description
Forest inventory is constantly looking to increase the precision and accuracy of the measured attributes while reducing measurements costs. A plethora of the tools can be used for measurements, out of which remote sensing provides the most cost-efficient way of estimating forest resources for large areas. Among the remote sensing data used for forest inventory, point clouds seems to be most suited for dimensional estimates, as they have a 3D structure. The strength and weakness of the point clouds used in forestry is the large amount of data describing the vegetation and the ground. The strength lies on the wall-to-wall coverage of the forest ecosystem, which is also its weakness, as most estimates are based on algorithms, that naturally are erroneous.
Modern forest inventory moved from stand level attributes to comprehensive description of individual trees from a stand. To capture the vertical dimension of each tree laser scans are currently used. The project will use laser scanner installed on unmanned aerial systems (UAS) and on handheld systems (HLS) to describe the forest from below and above the canopy. Using fused terrestrial and aerial point clouds, a forest inventory will be executed using deep learning algorithms training using either point clouds or rasters (images) created from point clouds.
The project has three objectives:
1. Convert raw data recorded by the sensors into a usable file format
2. Fuse aerial and terrestrial point clouds
3. Create the input for the deep learning algorithms by segmenting point clouds and rasters
Modern forest inventory moved from stand level attributes to comprehensive description of individual trees from a stand. To capture the vertical dimension of each tree laser scans are currently used. The project will use laser scanner installed on unmanned aerial systems (UAS) and on handheld systems (HLS) to describe the forest from below and above the canopy. Using fused terrestrial and aerial point clouds, a forest inventory will be executed using deep learning algorithms training using either point clouds or rasters (images) created from point clouds.
The project has three objectives:
1. Convert raw data recorded by the sensors into a usable file format
2. Fuse aerial and terrestrial point clouds
3. Create the input for the deep learning algorithms by segmenting point clouds and rasters
Describe the type of work and tasks you anticipate the student will perform
The duties are sequential, in the sense that first the raw data will be converted into the standard point cloud format, then the above and below canopy point clouds would be merged, and lastly, the training data for the deep learning algorithms will be created.
The task needed to produce the training data for the deep learning forest inventory algorithms are:
• Execute terrestrial laser scanning
• Execute ground inventory using hypsometers and D-tape
• Process raw data in DJI Terra and FJ Triton
• Delineate tree crowns on aerial images
• Label points within point clouds using QTM
The task needed to produce the training data for the deep learning forest inventory algorithms are:
• Execute terrestrial laser scanning
• Execute ground inventory using hypsometers and D-tape
• Process raw data in DJI Terra and FJ Triton
• Delineate tree crowns on aerial images
• Label points within point clouds using QTM
Please list special or preferred skills
To perform the required tasks, the student should be familiar with forest inventory and GIS (any software).
The student will execute at least 5 trips o the McDonald -Dunn forest where he will stem map several plots (more than 50 plots). One characteristic of this work is that the student will spend time also in front of the computer processing images and analyzing 3D objects,.
The student will execute at least 5 trips o the McDonald -Dunn forest where he will stem map several plots (more than 50 plots). One characteristic of this work is that the student will spend time also in front of the computer processing images and analyzing 3D objects,.
Hourly rate of pay
17
Certification
Yes
What is the expected timeline of this project?
Nov 1, 2025 to May 31, 2026
Are special skills or knowledge required to work on this project?
Yes
Will training be provided?
Yes
How many hours per week do you anticipate a student to work?
20
How many hours per week do you anticipate engaging in direct mentorship?
4
To ensure the successful completion of the project, understood not only by its deliverables, but also by the interaction between mentor and mentee, I will dedicate a minimum of 2.5 hours / week to discuss the project and various facets of scientific investigation. At least 45 min/week will be dedicated for assessment of project’s progress, specifically parameter selection and algorithms implementation. Additionally, approximately 45 min/week will be spent discussing issues, possible solutions, and ensuring rigor to the project.
I will train the student in using commercial software QTM, Geoslam Connect, FJ Triton, and DJI Terra.
I will train the student in using commercial software QTM, Geoslam Connect, FJ Triton, and DJI Terra.