Date Available

1-3-2022

Year of Publication

2022

Degree Name

Master of Science in Biosystems and Agricultural Engineering (MSBiosyAgE)

Document Type

Master's Thesis

College

Engineering

Department/School/Program

Biosystems and Agricultural Engineering

First Advisor

Dr. William Ford

Abstract

The need for assessment and management of aquatic vegetation in stream ecosystems is recognized given the importance in impacting water quality, hydrodynamics, and aquatic biota. However, existing approaches to monitor are laborious and its currently not feasible to track spatial and temporal differences at broad scales. The objective of this study was therefore to map and classify aquatic vegetation of a shallow stream with heterogenous mixtures of emergent and submerged aquatic vegetation. Data was collected in the Camden Creek watershed within the Inner Bluegrass Region of central Kentucky. The use of unmanned aerial vehicles (UAVs) was employed and both visible (RGB) and multispectral imagery were collected. Machine learning techniques were applied in an off-the-shelf software (QGIS environment) to develop visible and multispectral classification land-cover maps following an effective object-based image analysis workflow. Visible images were additionally coupled with high frequency water quality data to examine the spatial and temporal behavior of the aquatic vegetation. Results showed high overall classification accuracies (OA=83.5% for the training dataset and OA=83.73% for the validation dataset) for the visible imagery, with excellent user’s and producer’s accuracies for duckweed, both for training and validation. Surprisingly, multispectral overall accuracies were substantial (OA=77.8% for the training dataset and OA=70.2% for the validation dataset) but were inferior to the visible classification results. User’s and producer’s accuracies were lower for almost all classes. However, this approach was unsuccessful in detecting, segmenting and classifying submerged aquatic vegetation (algae) for both datasets. Finally, a change detection algorithm was applied to the visible classified maps and the changes in duckweed areal coverage were successfully estimated.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2022.034

Funding Information

This study was supported by the National Science Foundation.

The award number was 1632888 and the project duration was from 2016-2021.

Share

COinS