Repository logo
 

Enhancing Community Conservation Efforts for Pest Plant Eradication Using Low-Altitude Remote Sensing

Files

Thesis(16.13 MB)

Date

2022

Authors

Al-Hili, Sarah Hekmet Hekmet

Supervisor

Gillman, Len
Bollard, Barbara
Leuzinger, Sebastian

Item type

Thesis

Degree name

Master of Science (Research)

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

Invasive plant species in New Zealand are a significant threat to biodiversity. These invasive plants were introduced during earlier settlements via different pathways. Regional councils and conservation groups attempt to these rebuild native flora and fauna communities by restoring, reducing, and halting human environmental impacts including the spread of invasive plants. The current method to detect invasive plants involve on-ground searches, which are slow, infrequent, and costly. Invasive plants require early, rapid detection and management. Remote sensing in modern literature has been used as a novel way of detecting invasive plants in remote areas. This method is favoured as it is faster, more cost-effective, and thus able to be performed regularly in contrast with current methods used. This study detected the invasive plants Woolly Nightshade, Moth Plant, Gorse, and Sweet Pea Shrub using low-altitude high-resolution drones on Moturoa Island in the Bay of Islands, New Zealand. A Phantom 4 Professional UAV paired with a MicaSense RedEdge Sensor was used to collect the imagery. To detect the invasive plants, we used three classification techniques in ArcGIS: pixel-based Maximum Likelihood, object-based Support Vector Machine and Random Trees. We tested nine segmentation parameters for each classier. Out of the nine segmentation parameters, parameters 2 and 8 performed the best for Random Trees and Support Vector Machine with an average of 60% and Kappa of 54%. However, Maximum Likelihood did not perform adequately with segmentation parameters 2 and 8; instead, it performed best with segment 9. The accuracy results for that parameter were 56%, with a kappa value of 50%. This research examined the need for a suitable sampling size, effects on accuracy in relation to the timing of imagery acquisition, segmentation settings and the effects of shadows on plant spectral reflectance. This research can directly benefit the Northland Regional Council, community-led restoration projects, and researchers in this field efficiently by enabling prompt detection of invasive plants.

Description

Keywords

Source

DOI

Publisher's version

Rights statement

Permanent link

https://hdl.handle.net/10292/15241

Collections

Masters Theses