Whiteside, Tim (2011) Multiscale vegetation characterisation of tropical savanna using object-based image analysis. ["eprint_fieldopt_thesis_type_phd" not defined] thesis, Charles Darwin University.
Restricted to Repository staff only
Available under License Creative Commons Attribution No Derivatives.
Download (10MB) | Request a copy
This thesis applies object-based image analysis (OBIA) to mapping spectrally variable land cover from moderate to high resolution satellite imagery. The study was undertaken over a 2600ha area within tropical northern Australia. The region is dominated by typical savanna vegetation characterised by continuous grass cover and discontinuous woody overstorey. The first objective examines the advantages of OBIA over per-pixel methods for mapping land cover. A comparison found object-based image analysis to be statistically superior (z=2.285 (p=0.01), McNemar’s χ2=8.966 (p=0.0028)). The second objective developed a rule-set for land cover classification of QuickBird data. For a subset of the study area the overall accuracy was 94% and K^ = 0.92. Applied to the entire area, accuracies were lower with error associated with burnt vegetation. The third objective investigated mapping vegetation structural attributes using OBIA. A tree crown extraction process was developed for QuickBird data. Accuracies over 75% were obtained, despite savanna Eucalypts exhibiting canopy characteristics hindering delineation. The fourth objective compared canopy cover estimates from extracted tree crowns to pixel-based and manually derived methods. Tree crown cover shows relationships with vegetation indices from QuickBird (r2=0.93) and ASTER (r2=0.22) imagery, and manually interpreted estimates from aerial photographs (r2=0.43). The final objective implemented area-based measures quantifying the spatial and thematic accuracy of OBIA. Results show these measures provide valuable thematic and geometric accuracy information provided appropriate reference data are available. This study has demonstrated OBIA is suitable for mapping land cover in spectrally variable landscapes at multiple scales. More specifically, OBIA has better accuracy over per-pixel methods, transferrable rule sets can be used to map land cover from high spatial resolution data, and OBIA methods can extract dominant vegetation structures. Finally, limitations of site-specific accuracy assessments can be addressed through area-based accuracy measures.
|Item Type:||Thesis (["eprint_fieldopt_thesis_type_phd" not defined])|
|Field of Research:||05 Environmental Sciences > 0501 Ecological Applications > 050104 Landscape Ecology
08 Information and Computing Sciences > 0801 Artificial Intelligence and Image Processing > 080106 Image Processing
09 Engineering > 0909 Geomatic Engineering > 090905 Photogrammetry and Remote Sensing
|Subjects:||G Geography. Anthropology. Recreation > GB Physical geography
G Geography. Anthropology. Recreation > GE Environmental Sciences
|Date Deposited:||19 Oct 2011 22:24|
|Last Modified:||20 Oct 2011 02:46|
Actions (login required)