Date of Award

2024

Degree Type

Dissertation

Degree Name

Doctor of Philosophy in Biological and Environmental Sciences

Specialization

Environmental and Earth Sciences

Department

Natural Resources Science

First Advisor

Jason Parent

Second Advisor

Andrew Davies

Abstract

Forests have immense economic value, contributing to industries such as timber production, nontimber forest products, and ecotourism (Wang et al., 2022). Additionally, forest ecosystems play a crucial role in mitigating climate change by acting as carbon sinks (Yingchun et al., 2012). However, forests also pose hazards to humans and infrastructure. Fallen trees, limbs, and contact of tree branches with power lines create numerous hazards to public safety, property damage, and substantial economic losses from power outages. Studies report that the total cost of storm-related power outages in the United States range from $20-150 billion annually (Campbell et al., 2021; Swaminathan et al., 1998; Eto et al., 2001), with about 40-85% of storm-related outages caused by trees (Simpson et al., 1996; Guggemos et al., 2003).

Monitoring forest areas with a high risk of mortality can guide management activities aimed at reducing hazards to infrastructure and public safety (Sonti et al., 2015). There is already a diverse assortment of remote sensing and GIS-based sensors, models, and applications that can be applied to the problems of forest management and monitoring (Somers et al., 2012; Kuenzer et al., 2011; Joshi et al., 2015; Kerr et al., 2003; Wulder et al., 2006; Wang et al., 2010; Hansen et al., 2013; Asner et al., 2015). However, there are gaps in the literature on detection and management of forest-related problems, such as the limited number of studies that use empirical data to evaluate existing management strategies for reducing power outages due to tree damage. Additionally, there is limited research exploring broad-scale tree mortality modeling in temperate deciduous forests and mapping individual trees using semi-automatic approaches.

My research aims to help fill these knowledge gaps by utilizing remote sensing and GIS techniques in forest monitoring and management related to human and infrastructure risks. The overall objective of this dissertation is to guide forest management in reducing tree-related power outages and hazards by evaluating management strategies and modeling forest mortality. The specific objectives were to:

  • Evaluate the effects of conductor coverings, enhanced tree trimming, and line characteristics on tree-related power outages.
  • Model tree mortality resulting from Spongy moth outbreaks using Landsat satel-lite imagery and GIS-derived environmental data.
  • Evaluate Deep Learning and ISODATA unsupervised classification for mapping dead trees using high-resolution aerial imagery.

My first chapter evaluates electric utility solutions to reducing tree-related power outages, specifically, the use of covered conductors and enhanced tree trimming (ETT). Tree-related outages during storms, from 2013 to 2017, were analyzed for power line segments. Covered conductors, ETT, and phase characteristics (single vs. multiple) were considered as treatments. To control for confounding factors, the treatment line segments were paired with control segments that have similar characteristics (i.e. percent canopy cover, line length, close proximity, etc.) but differed in their combination of treatments. Covered conductors and ETT were similarly effective in reducing power outages with median rate reductions ranging from 0.19 to 0.29 outages/km/year. When used together, covered conductors and ETT exhibit a significant additive effect, resulting in median reductions of 0.29 to 0.33 outages/km/year compared to bare conductors with no ETT. Moreover, multiphase power lines show lower median outage rates compared to single-phase power lines, suggesting that phase characteristics could serve as a useful predictor of storm-related outages.

The second chapter of my dissertation focuses on modeling forest mortality resulting from a Spongy moth outbreak in the temperate deciduous forests of Rhode Island between 2015 and 2017. Landsat-based defoliation mapping and geospatial environmental data were used with Random Forest to predict the severity of canopy tree mortality at a 100 m spatial resolution. Defoliation mapping was conducted by comparing mid-summer Normalized Difference Vegetation Index (NDVI) during outbreak years to baseline mid-summer NDVI from pre-outbreak years. A total of 21 predictors were used including geospatial data representing soil characteristics, drought conditions, forest characteristics, and proximity to coast, development, and water bodies. Forest mortality models achieved overall accuracies of 82% and 65% for the two-class (i.e. low/high mortality) and three-class (i.e. low/medium/high mortality) models, respectively. Key predictors of forest mortality included defoliation severity, distance to the coast, and canopy cover. Repeated defoliations were infrequent during the outbreak, and model performance only slightly improved with the inclusion of more than three variables. The models classified 35% of forests as experiencing canopy mortality exceeding 5 trees per hectare and identified 21% of Rhode Island forests with mortality exceeding 11 trees per hectare.

The third chapter of my dissertation mapped tree mortality at the individual tree scale directly from summer aerial imagery. The study area was located in Rhode Island's temperate deciduous forests following the Spongy moth outbreak in 2015-2017. Three common image classification methods - pixel- and object-based unsupervised classification, and Deep Learning - were evaluated for dead tree detection. The methods were applied to summer 2019 true-color aerial imagery with 7.6 cm resolution. The pixel-based classification achieved the highest accuracy (F1=0.84), and Deep Learning had the lowest accuracy (F1=0.67). Spectral and morphological filtering substantially reduced commission error for both the pixel- and object-based methods. The pixel-based method had the advantages of being automatable and not requiring training data. The study demonstrates the potential for image classification to extract dead deciduous trees from aerial imagery.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.