Back to search

University of Nevada - Reno

Phenology-based UAV remote sensing for classifying invasive annual grasses to the species level

Abstract

dc:description.abstract

The spread of invasive plant species severely alters wildfire regimes, degrades critical habitat for native species, and has detrimental impacts on ecosystem function, rangeland productivity, and long-term carbon storage dynamics. Remote sensing technology has greatly improved our understanding of invasive plant ecology and ability to map and monitor plant invasions. Mapping plant invasions to the species level with conventional satellite and airborne data has proven challenging, however, because many invasive species occur at fine spatial scales or are mixed with native species, and satellite passes may occur too infrequently to capture important phenological stages. Imagery derived from readily deployable Unmanned Aerial Vehicles (UAVs) offers high-resolution data over carefully timed acquisition dates during the growing season. However, some challenges remain that are particular to high spatial resolution imagery, where excessive detail from shadows and canopy gaps often result in misclassification, inaccuracy, and a "salt-and-pepper" effect in the final classification. The addition of textural and vegetation height data to a purely spectral pixel-based approach has the potential to mitigate these challenges and improve species-level vegetation classification. Using UAV imagery acquired at specific phenological stages, we investigate which combinations of spectral, textural, vegetation height, and multi-temporal techniques best separate two invasive annual grasses, cheatgrass and medusahead, to the species level.We selected five study sites ranging in area from 8 to 36 hectares (ha) in Paradise Valley, Nevada, which feature a variety of invasive and native species that are typical of the Great Basin region. For three carefully selected dates over the growing season during which cheatgrass and medusahead were most spectrally distinct, we conducted UAV flight campaigns and collected field data on vegetation composition. Imagery was processed in photogrammetric software to produce orthomosaics, digital terrain models, and digital surface models from which vegetation height was derived. Texture analysis was performed over the acquired raster data products. Multi-date spectral, textural, and vegetation height variables were used to predict vegetation class type using Random Forest machine learning methods. The overall goal of this research is to further remote sensing methods for vegetation classification of invaded landscapes to the species level. We investigated which combinations of spectral, textural, vegetation height, and multitemporal techniques best separate two invasive annual grasses - cheatgrass and medusahead. To explore the impact of explanatory variables in our classification, all possible additive combinations of our variables were calculated. We found that multi-temporal texture variables and vegetation height added additional levels of information to our classification and, when combined with multi date spectral information, achieved the highest overall accuracy. Our model resulted in a robust classification across several diverse study sites.

Degree

thesis:*
Level thesis:degree_level
Master's Degree
Department thesis:degree_department
Environmental and Natural Resource Sciences
Year dc:date.issued
2021

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Ready, Alice Anne
Advisor dc:contributor.advisor
  • Weisberg, Peter J
Committee members dc:contributor.committeemember
  • Greenberg, Jonathan A
  • Nussear, Kenneth E

Subjects

dc:subject × 6

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/11714/7971

Chain of custody

source
Harvested from
University of Nevada - Reno
Base URL
scholarwolf.unr.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Ready, Alice Anne. Phenology-based UAV remote sensing for classifying invasive annual grasses to the species level. Master's Degree thesis, 2021. http://hdl.handle.net/11714/7971