Into the canopy: advancing forest resource characterizations with lidar

Blackburn, Ryan Christopher (2022) Into the canopy: advancing forest resource characterizations with lidar. Doctoral thesis, Northern Arizona University.

Text Blackburn_2022_into_canopy_advancing_forest_resource_characterizations.pdf - Published Version Download (5MB)

Abstract

Forests around the world are facing radical shifts in distributions and disturbances resulting from climate change that are exacerbated by anthropogenically caused novel uncharacteristic forest conditions. This is especially true in the western US where shifts in structure and composition have resulted in denser (i.e., more trees), fire-intolerant and tree-dominated ecosystems with reductions in function and ecosystem services. To address this, ecological restoration has been proposed across millions of hectares with the goal of shifting forest structure and composition closer to its historical range of variability (HRV). However, forest restoration necessarily alters existing, degraded forest conditions – conditions wildlife have adapted to occupy - and while increased function and health is known, impacts to habitat and resulting animal behavior is still being studied. This knowledge gap has created a need to quantify present forest conditions at a higher fidelity and at landscape scales to develop a deeper understanding of restoration practices and wildlife habitat interactions. In this dissertation, I investigated recent developments in methodologies using light detection and ranging (lidar) and machine learning to create workflows to better assess forest conditions. I used three separate studies conducted in the Southwest to validate methods for quantifying forest structure and composition and to provide insights into spatially-explicit conditions around Mexican Spotted Owl (Strix occidentalis lucida; MSO) nesting habitat. To quantify forest structure, I compared three different approaches (i.e., area-based, tree-based, voxel-based) for estimating forest attributes (i.e., basal area, volume, biomass) across 1,680 field plots in Arizona and New Mexico using random forests and ridge regression. Boruta feature selection was performed on variable subsets, including a mixture of all lidar-derived predictors (my application of Boruta using the “Kitchen Sink”, or KS-Boruta). A corrected paired t-test was utilized to compare six validated models (area-Boruta, tree-Boruta, voxel-Boruta, KS-Boruta, KS-all, ridge-all) for each forest attribute. Based on significant reductions in the symmetrical median absolute percent error, basal area and biomass were best modeled with KS-Boruta, while volume was best modeled with KS-all or the “kitchen sink” using voxel, tree and area-based approaches. Analysis of variable importance showed voxel-based predictors are critical for the prediction of the three forest attributes. This study highlights the importance of multi-resolution voxel-based variables for modeling forest attributes in an area-based context. To assess forest composition, I used deep learning techniques to create a classification model for seven species from lidar imagery within the Mogollon Rim Ranger District of the Coconino National Forest. I compared individual tree segmentation images of unmanned aerial vehicle laser scanning (UAV-LS) and airborne laser scanning (ALS) data colored by different eigenvalues in a multi-view 2D convolutional Neural Network. The final models for each acquisition type were curvature-colored ALS imagery (88% accuracy) and verticality-colored UAV-LS imagery (88% accuracy). This highlights the capability of species classification across landscapes using ALS data alone. Finally, I use lidar to assess forest conditions over varying spatial extents around MSO nesting habitat. I explored changes over varying spatial extents in 137 lidar-derived forest attributes including the voxel metrics found important in my first manuscript chapter. I focused primarily on canopy strata, gap sizes, snags, the mean frequency ratio voxel-based variable, and the coefficient of variation for the mean percentage of points greater than two meters above the ground. The goal of the study was to determine which forest attributes were most strongly associated with MSO nesting habitat by assessing 10-meter annuli around owl nests and manager-delineated owl habitat areas such as cores and protected area centers (PACs). Additionally, I wanted to determine if there were differences between manager-delineated PACs and circular buffered areas of 243 hectares (i.e., recommended minimum PAC size; P880) in the immediate vicinity of the nest tree. I explored niche overlap between each owl habitat area and between each annulus and the P880 region. Kruskal-Wallis and post-hoc Dunn’s tests were performed across regions to determine significant differences. Many of the forest attributes had differing niche overlaps between core and PAC. However, PAC and P880 comparisons did not have as many differing attributes. Kruskal-Wallis results also confirmed that some variables differed between core and PAC regions, but not PACs and P880s. Tall trees (18-34 meters) and the mean frequent ratio were important indicators of MSO nesting habitat structure across all analyses. This study indicated that owl habitat is a more nuanced structure and that increased forest cover canopy cover (where the cover of tall trees is commonly found) is an inadequate description. Recently occupied Mexican Spotted Owl habitat might best be characterized as areas with higher canopy cover, but also with tall trees and open sub-canopies. This dissertation provides building blocks on a larger body of literature pushing forest management and monitoring into a new technological age where comprehension of fine-grain attributes over landscape-scale extents is no longer out of reach.

Item Type:	Thesis (Doctoral)
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:	forest structure; lidar; Mexican Spotted Owl; neural network; species composition; voxel
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:	14 Jul 2022 17:22
Last Modified:	25 May 2023 08:30
URI:	https://openknowledge.nau.edu/id/eprint/5848

Actions (login required)

IR Staff Record View

Downloads