Using satellite imagery and novel low altitude aerial imagery to classify coastal wetland vegetation for change detection at Whatipu Scientific Reserve, Auckland, NZ

• Main Author: Lawrence, Grant (Author)
• Other Authors: Breen, Barbara (Contributor), Bishop, Craig (Contributor)
• Format: Others
• Published: Auckland University of Technology, 2016-05-08T23:54:55Z.
• Subjects: Remote sensing; GIS; Unmanned aerial systems; Vegetation; Wetland; Supervised classification; Satellite imagery; Thesis
• Online Access: Get fulltext

 Wetland vegetation mapping is an important technical task for managing and maintaining essential ecosystem services that wetlands provide. Despite their importance, wetland ecosystems are highly threatened in New Zealand with less than 10% of pre-human extent remaining. Remote sensing has advantages over traditional techniques, allowing non-destructive sampling of resources and enabling users to gain critical information more quickly and cheaply. The potential for remote sensing to provide an increased understanding of coastal wetland environments has not been realized in New Zealand. The collection and satellite simulation of spectral data for 14 species at Whatipu Scientific Reserve provides valuable information for the application of imagery classification and for future research. Despite low spectral separability between these species, a relatively accurate land cover map was established for each of the multi-date satellite imagery sets, with individual class accuracies between 75% and 99%, depending on vegetation type. This indicates that high-resolution multispectral imagery (2m spatial resolution) such as WorldView 2 and 3 satellite imagery products show good potential for the identification and classification of coastal wetland vegetation. In addition, although satellite remote sensing platforms are useful for vegetation mapping they still require field training and validation samples. This study investigated the use of low altitude Unmanned Aerial System (UAS) imagery (6cm spatial resolution) for the collection of training and validation data crucial for the classification of multispectral satellite imagery. By using ancillary data and UAS imagery, I minimised the need for extensive field surveys that are potentially destructive, timely and expensive. The land cover changes determined from the multi-date classifications at Whatipu show minimal change in the past 4.5 years, however, changes that were detected are significant, particularly with the expansion of exotic shrubland species. The high-resolution UAS imagery also provided sufficient detail to accurately identify exotic Pampas (Cortaderia Selloana) in comparison to high-resolution (36cm spatial resolution) satellite imagery products.