Methods for mapping the tundra vegetation of sub-Antarctic Macquarie Island

• Main Author: Bricher, PK
• Format: Thesis
• Language: English
• Published: 2012
• Subjects: image classification; plant communities; species distribution modelling; terrain analysis; geographic stratification; random forests; Antarctic; Antarc*; Macquarie Island; Tundra
• Online Access: https://eprints.utas.edu.au/14744/; https://eprints.utas.edu.au/14744/1/Bricher_whole_PDF_thesis.pdf

Vegetation mapping is increasingly used for detecting changes in plant distributions at landscape scales. This is aided by the development of sophisticated machine learning tools and very high resolution satellite imagery, which together allow for the detection of changes at finer spatial and attribute resolutions. Even with these new tools, the sub-Antarctic poses particular challenges for mapping. The remoteness demands particular efficiency in field sampling and regular cloud-cover hampers satellite image acquisition and the tundra vegetation is small-statured and dominated by ecologically generalist species. Additionally, High accuracy mapping is of particular interest on sub-Antarctic Macquarie Island, where the vegetation is changing rapidly. Reliable maps of current species distributions would provide valuable baselines for monitoring future change. This thesis develops and tests methods for all steps of the mapping process, including field methods, sampling design, examining whether plant communities are an appropriate mapping unit for change detection, and testing the best methods of image classification. A new photographic field sampling method is demonstrated to collect equivalent data to an existing field quadrat method but with less field time needed. Next, a geographically-stratified random sampling design is developed, applied and tested for its capacity to capture the variation in terrain variables that are thought to drive plant species distributions. Previous descriptions of the vegetation of Macquarie Island are based on the idea that the vegetation can be more or less grouped into discrete communities. I test this hypothesis and find that floristic classification is unable to identify any stable communities on this highly disturbed island. I therefore finally focus on single-species mapping, determining the best combination of very high resolution satellite imagery, terrain modelling and random forest classification to map the extent of the endemic, critically-endangered cushion plant, Azorella macquariensis, with very high accuracy. The successful mapping of A. macquariensis, a small-statured and patchily distributed species, demonstrates the utility of this mapping technique. I describe initial results for applying this technique to other species, highlighting the ecological and spectral characteristics that distinguish those tundra species that are amenable to mapping from VHR satellite imagery and terrain modelling. In summation, in this thesis, I develop and test tools for all stages of mapping the distribution of plant species on sub-Antarctic Macquarie Island at a resolution suitable for monitoring rapid distributional changes.