Date of Award

June 2019

Degree Type


Degree Name

Master of Science (MS)


Earth Sciences


Christa Kelleher


Drone, Land Cover Classification, Linear Discriminant Analysis, Remote Sensing, UAV

Subject Categories

Physical Sciences and Mathematics


For decades, remote sensing has been used by scientists and planners to make detailed observations and decisions on areas with industrial problems, remediation and development sites, and resource management. It is challenging to make high spatial and temporal resolution observations along headwater and small streams using traditional remote sensing methods, due to their high spatial variability and tendency for rapidly changing water quality and discharge. With improved technology in sensors and launching platforms, remote sensing via Unoccupied Aerial Vehicles (UAVs) now allows for imagery to be collected at high spatial and temporal resolution, with the goal of providing a deeper analysis of these intricate and difficult to access regions. One recent area of interest is the use of UAVs to delineate land and water cover. While recent innovations in low-altitude multispectral and hyperspectral imagery have been used extensively for tracking land cover, it has been used less frequently to detect changes within the water column through space and time. In addition, it is unclear whether classification methods applied to headwater systems are translatable across adjacent stream reaches or across flights on different days, as well as how much information is needed to perform such classifications. This study demonstrates that UAV multispectral imagery can be used to classify land cover as well as uniquely identify submerged aquatic vegetation by combining methods of remote sensing, image processing and machine learning. A linear discriminant analysis (LDA) model was developed to provide land and water cover classification maps (with statistical analysis of error) using training data from hand delineated multispectral shapefiles. This method proved to be robust when classifying land cover along a single reach, even when using a very small proportion of the training data. Through attempts to transfer data through space and time, this exercise highlights the shortcomings in multispectral imagery and the dependence on lighting conditions, reach orientation and shading from nearby structures such as vegetation. Therefore, this approach is likely most beneficial for classifying land cover and submerged aquatic vegetation at a single reach for a single time, but more work must be done to further identify physical limitations of multispectral imagery and calibration methods which might allow for an “absolute” measure of reflectance.


Open Access



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.