Using multi-platform remote sensing methods to map chenopod shrubland communities in Witchelina Nature Reserve in SA Arid Lands

Author: Yun Guang Jasper Wong

Wong, Yun Guang Jasper, 2021 Using multi-platform remote sensing methods to map chenopod shrubland communities in Witchelina Nature Reserve in SA Arid Lands, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

Environmental monitoring over a large expanse in remote areas is difficult with traditional in-situ vegetation surveys. Remote sensing helps to overcome the problem by balancing the need to monitor a large study area with adequate spatial resolution to detect the object of interest across a large area. This study uses high-resolution 0.5 metre Pleiades satellite imagery, together with imagery from aerial surveys at 0.03m spatial resolution taken with a camera mounted on an Unmanned Aerial Vehicle (UAV) which is acting as pseudo ground control. Classification of the Pleiades imagery is being used to map chenopod shrubs with 4 distinct spatial distribution patterns in Witchelina Nature Reserve located in the western Flinders Ranges in the South Australian (SA) arid lands. The objective is to employ machine learning methods to achieve mapping of the chenopod habitat of the endangered Thick-bill Grass Wren from the high-resolution satellite imagery. A comparison will be made between two image classification methods- Image Segmentation and classification using Support Vector Machine (SVM) and Deep Learning using convolutional neural network (CNN). Accuracy will be assessed using withheld pseudo ground truth data and suitability for objective. If the classification accuracy is sufficiently high enough, the cost of acquiring and processing satellite imagery will be computed for the whole of the Witchelina Nature Reserve.

Keywords: remote sensing, machine learning, image classification, witchelina nature reserve, mapping vegetation, mapping, classification, maps, satellite imagery, drone, arcgispro, deep learning, CNN, segmentation, image segmentation, supervised classification, unsupervised classification, high-resolution satellite imagery

Subject: Geography thesis

Thesis type: Masters
Completed: 2021
School: College of Science and Engineering
Supervisor: David Bruce