Repository logo
 

Predictive models of individual tree health: the utility of uncrewed aerial system data to inform forest management

Abstract

Informed forest management is an essential tool for increasing forest resilience as global climate change increases the frequency and severity of disturbances. Drought and fire are inherent components of natural forest ecosystems, yet the increase in their occurrence combined with increasing severity threatens forest recovery pathways and their ability to sequester carbon. Fire is an essential disturbance that regulates both local and global ecosystem processes and functions. In forests of the United States, fire exclusion during the 20th century drove increases in vegetation density and connectivity, while altering ecosystem composition and structure. As a result of this increase in forest density and fuels, the U.S. has experienced increases in the severity, size, and frequency of wildfires, and drought. Fuels treatments in the form of mechanical thinning and prescribed fire provide opportunities to reduce forest density and fuels, research treatment effectiveness, and reduce the risk of catastrophic fires. Remote sensing allows researchers to collect spatially continuous data on forest conditions both before and after treatments, to study treatment-based impacts. Uncrewed aerial systems (UAS), specifically, provide a tool with user-controlled temporal and spatial resolutions. These systems enable the collection of multispectral and three-dimensional data, via structure from motion (SfM) or LiDAR, allowing researchers to examine changes to structure, composition, and health at ultra-high-resolution. The research within this dissertation focuses on combining pre- and post-treatment multispectral and structural data to assess individual tree health and changes to that health induced by disturbances. Chapter 2 investigates the utility of band-equivalent reflectance (BER) data from a consumer-grade multispectral UAS camera to predict sapling drought stress in western white pine (Pinus monticola) and Douglas-fir (Pseudotsuga menziesii var. glauca) through a controlled laboratory experiment. The results demonstrate that BER data can reliably detect physiological stress under controlled conditions, providing a strong foundation for using low-cost remote sensing to monitor tree health. Chapter 3 evaluates the transferability of the lab-developed BER models to mature ponderosa pine (Pinus ponderosa) and Douglas-fir forests in northern Colorado. This chapter evaluates the models' ability to predict relative drought stress in natural forests and explores the development of site-specific models to improve accuracy. Results reveal that while lab-based models provide valuable insights, field-developed models significantly enhance predictive performance, emphasizing the importance of adapting methodologies to specific field conditions and species. Chapter 4 quantifies the impact of compounding drought stress and fire, of varied intensities, on sapling physiology and mortality for western white pine and Douglas-fir. This chapter modeled the interaction of sapling drought stress and fire intensity on mortality using a dose-response strategy and pre-fire physiological and morphological characteristics. Results demonstrate that increasing drought stress results in higher post-fire mortality compared to less-stressed saplings subjected to the same fire intensity, regardless of species. Chapter 5 examines our ability to predict individual tree crown volume scorch in longleaf pine (Pinus palustris) forests treated with prescribed fire. This chapter assesses the impacts of data collection timing, inclusion of pre-fire data, and spectral range on model accuracy. The results of this chapter demonstrate that crown volume scorch can be successfully modeled using post-fire imagery as soon as one-day post-fire, with any sensor that includes red-green-blue data. This collection of research advances our understanding on the use of UAS data for modeling forest health. The models of relative drought stress could be integrated with forest structure and composition metrics to inform site-based thinning to optimize post-treatment stand resilience. Further, scorch classification models could be used to examine patterns of fire effects, providing critical feedback about prescribed fire ignition patterns. These models have the potential to be integrated into operational forestry practices and can provide actionable insights to guide management that promotes forest resilience and disturbance recovery.

Description

Rights Access

Subject

fire ecology
machine learning
crown volume scorch
UAS
forest health

Citation

Associated Publications