Year of Award

2022

Document Type

Thesis

Degree Type

Master of Science (MS)

Degree Name

Forestry

Department or School/College

W.A. Franke College of Forestry and Conservation

Committee Chair

Carl Seielstad

Commitee Members

Lloyd Queen, Charles Palmer

Keywords

UAS, fire, drone, fuels, behavior, qaudcopter

Subject Categories

Forest Management

Abstract

Understanding how fuel, weather, and terrain interact to produce fire behavior continues to motivate fire science andhas resulted in development of new physics-based fire behavior models that place increased demands on input data such as fuels. Recent technological advancements in computing, unmanned aerial systems (UAS), and sensors (RGB, multispectral, thermal, and hyperspectral cameras) can provide new opportunities for land managers and scientists to advance knowledge of fuels and fire behavior and their interactions on the landscape. In this study, imagery from high resolution multispectral cameras mounted on UAS were used to build orthomosaics and point clouds of surface fuelbeds in grass, litter, and shrub fuels of the Sycan Marsh Preserve in Oregon. The purpose of this effort was to develop useful inputs to a fuels translator called STANDFIRE that prepares fuels data for use in physics-based fire models. Fuel type polygons were delineated using traditional photo-interpretation for nine 1 ha plots that were ultimately treated with fire. Each fuel polygon was attributed from field-collected data based on their dominant fuel type. Differences between fuel type polygons were assessed statistically to document the distinctiveness of each fuel type, to overcome field sample-size limitations, and to provide logic for merging fuel types that were similar. Additionally, 3D point clouds and orthomosaics were examined to better understand their information content for more detailed characterizations of fuels. In this latter part of the research, shrub height, width, and cover were extracted from the point clouds and compared to field measurements. The findings were as follows: Defensible fuel type classes were easily delineated using photo-interpretation, resulting in 21.4% of the cumulative plot area classified as litter, 65.3% as grass and 10.3% as shrub fuels. Effective attribution of fuel polygons was dependent on how and where field data were collected and differed by year. Lack of sufficient sample sizes in some fuel type polygons required aggregation of field data from all plots within the Brattain burn unit in 2018. These shortcomings were overcome in 2019 by acquiring rapid-look imagery prior to field sampling that enabled more balanced samples across the range of variability, along with utilization of precision GPS. Within the point-clouds, shrub height was underestimated while width was over-estimated. Shrub cover was also under-predicted from the point cloud and was better enumerated using a conventional dot-grid approach on the orthomosaic. Improvements in data collection methods from 2017-2019 have resulted in a stable workflow that produces consistent fuels data formatted for STANDFIRE. The polygon-based approach is suitable for use in fire model validation due to its ability to rationally integrate sparse field data, because STANDFIRE is designed to work easily with polygons, and because there is insufficient evidence that model validation is at a point where it will benefit from use of more complex pixel or object-based inputs. Automated approaches to polygon delineation via region-growing, machine learning, and segmentation are a logical next step, with the caveat that the inputs derived in this study should be tested in the modeling environment first.

Download

Share

COinS
 

© Copyright 2022 Matthew R. Cunningham