Derivation of a Perennial Vegetation Density Map for the Australian Continent
Multi-temporal Landsat coverages of the entire Australian continent have been assembled under the Australian Greenhouse Office (AGO) NCAS Land Cover Change Program. The major purpose of this program was the production of spatially detailed classifications of the extent and change in area of forest cover for Australia since 1972 for input to carbon modelling. Changes in vegetation density may also contribute significantly to the carbon budget, and are also of major significance in natural resource management. Landsat has frequently been used for the estimation of biophysical parameters, such as cover or crown density. Regression analyses of image and ground data have commonly been applied. Operational monitoring of vegetation trends from time series of cover/density indices has also been demonstrated in rangeland and forested environments.
The AGO imagery provided an opportunity to produce a Landsat-derived estimate of perennial vegetation density for the entire continent using a consistent methodology. Standard methods were defined to provide density ‘ground truth’ from Ikonos images acquired in 2002. Over 300 Ikonos images from AGO archives were used. Vegetation density was recorded for multiple sample areas within each. Ground and image data from multiple 1:1m mapsheets were analysed together. Broad stratification zones and scene boundaries were used. A variety of analysis techniques were examined, including linear models and ‘random forests’. Random forests is a powerful non-parametric tree ensemble technique. Methods and results were compared on the basis of ‘goodness of fit’ and comparison of output maps. Random forests performed better than linear models in a range of test areas, and this method was applied nationally.
The process has produced a map of vegetation density for Australia, limited at present to areas within the AGO forest mask. Details of the Random Forest procedure, the ground data, and the analysis will be discussed.