Classified earth observation data between 1990 and 2015 for the Perth Metropolitan Region, Western Australia using the Import Vector Machine algorithm

This dataset represents land cover for 7 sequential snapshots (1990, 2000, 2003, 2005, 2007, 2013 and 2015) over the Perth Metropolitan Region, Western Australia (WA) derived from medium resolution Landsat data. Cloud free imagery was acquired in or close to the month of July coinciding with WA's winter months coinciding with peak green-up facilitating the greatest contrast between spectrally similar surfaces (e.g. bare earth and urban). Imagery was first standardised and normalised to remove inherent residual noise (e.g. differences in modelled atmospheric correction parameters) whilst permitting classification of all imagery based upon a single classification model. The model was computed from the 2005 image representing the month post maximum rainfall of all considered imagery associated with peak greenness and maximum spectral separability. Classification of the normalised data was achieved with the Import Vector Machine (IVM) algorithm following a hybrid forward/backward strategy that adds import vectors whilst continuously testing validity in each step, producing a sparse and more accurate classification solution. Classified land cover data is provided in raster format (.tif) and divided into the classes: bare earth (1), grassland (2), low urban albedo (e.g. asphalt (3)), water (4), forest (5) and high urban albedo (e.g. concrete (6)). Please see MacLachlan et al. (2017) for further details.