Deep learning-based coarse woody debris biomass estimation from mobile lidar

Reiser, Merideth (2023) Deep learning-based coarse woody debris biomass estimation from mobile lidar. Masters thesis, Northern Arizona University.

Text Reiser_2023_deep_learning-based_coarse_woody_debris_biomass_estimation.pdf - Published Version Download (1MB)

Abstract

Coarse woody debris is essential to the structure and function of forested ecosystems. It plays a role in carbon and nutrient cycling, soil health, and hydrological processes, and it provides a heterogeneous substrate for the development of forest soils, wildlife habitats, forest regeneration, and fire processes. Accurate coarse woody debris loading values are necessary to model fire behavior and calculate carbon budgets. Line-intercept methods, specifically Brown’s transects, are the most commonly used methods for measuring coarse woody debris. While this method is fast, it is also labor intensive, especially in rough terrain and where access is limited. Brown’s transects also cannot capture the variability of coarse woody debris biomass within a forest stand, which is the scale at which most fuel treatments occur. Advancements in remote sensing have opened new avenues for measuring forest biomass at larger scales and with finer resolutions. Mobile Lidar scanning (MLS) collects high-resolution 3D structural data of forest plots, which enables actual volumetric calculations instead of relying on geometric approximations. This technology can potentially conduct forest structural inventories that are competitive with fixed-area sampling methods in the time it takes to collect Brown’s transects. However, this requires segmentation methods that distinguish vegetation from stems and coarse woody debris from the ground. The Forest Structure Complexity Tool (FSCT) is the first open-source program using deep learning to do just that. To assess this tool’s performance on coarse woody debris segmentation, which is the most notable weakness of FSCT, we compared the original model and a model that we retrained for our study site to Brown’s transect and fixed-area plot measurements in the dry mixed-conifer and ponderosa pine forests on Mogollon Rim, AZ. Linear models fit to the original, untrained model predictions and field observed coarse woody debris loads produced R2 values of 0.37 and 0.62 for ponderosa and dry mixed-conifer, respectively, and 0.38 and 0.45 for our retrained model, suggesting FSCT could be ready-to-use without the need for any user-defined parameters or tedious retraining steps. While Brown's transects overestimated mean coarse woody debris loads by 15.8% and 84.8% for ponderosa and dry mixed-conifer, respectively, FSCT underestimated these coarse woody debris biomass, with the original model underestimating by 19.5% and 10.7%, and the retrained model by 22.6% and 51.8%. We determined that for our study site, FSCT was able to provide estimates of coarse woody debris loading that aligned well with current field sampling methods. While FSCT needs to be tested and retained in more conditions, this tool may allow managers to utilize MLS technology for accurate coarse woody debris measurements.

Item Type:	Thesis (Masters)
Publisher’s Statement:	© Copyright is held by the author. Digital access to this material is made possible by the Cline Library, Northern Arizona University. Further transmission, reproduction or presentation of protected items is prohibited except with permission of the author.
Keywords:	CWD loading; mixed-conifer; MLS; ponderosa pine; restoration; watershed; Course woody debris; Forest Structure Complexity Tool; FSCT
Subjects:	S Agriculture > SD Forestry
NAU Depositing Author Academic Status:	Student
Department/Unit:	Graduate College > Theses and Dissertations College of the Environment, Forestry, and Natural Sciences > School of Forestry
Date Deposited:	01 Oct 2025 17:35
Last Modified:	01 Oct 2025 17:35
URI:	https://openknowledge.nau.edu/id/eprint/6202

Actions (login required)

IR Staff Record View

Downloads