Date of Award

Spring 2017

Degree Name

Master of Science (MS)

Department

Geosciences

Advisor

Dr. Thomas Schafer

Abstract

The interest in using unmanned aerial systems (UAS) for remote sensing of natural resources and ecology has grown rapidly in recent years and continues to develop. Recent improvements in the cost, size, and accessibility of consumer-grade UASs are now facilitating image collection on low-flying UAS platforms. Rangelands provide a unique opportunity to explore the uses of UAS remote sensing due to their large spatial extent and spatial and temporal heterogeneity. This study focuses on the use of UAS’s to identify rangeland vegetation in a southern mixed-grass prairie in Kansas. Two primary questions were asked: 1) Do small UASs with low-cost sensors collect imagery useful for mapping rangeland plants? 2) Do different supervised classification techniques yield significant differences in their ability to classify rangeland vegetation? Data were collected over 100 acres of the Hadley Range in north-east Ellis County. Three separate modeling algorithms (Maximum Likelihood Classifiers, Random Forests, and Support Vector Machines) were compared to a random dataset to determine if imagery collected with UAS could identify land cover better than random assignment. While the classification algorithms did perform better than random in most regards, they did not perform sufficiently well as to replace, or even compare to field work. However, I expect that as technical, and technological improvement in spatial, spectral, and temporal resolution occur, UAS remote sensing’s ability to aid in determining stocking rates for grazing, locating invasive or rare species, and estimating overall biodiversity will greatly increase.

Rights

Copyright 2017 Adam Rusk

Comments

Notice: This material may be protected by copyright law (Title 17 U.S. Code).

Included in

Geology Commons

Share

COinS