Estimating sub-pixel components of a semi-arid woodland

Abstract

Mixture models were applied to Landsat-MSS and -TM data in a semi-arid woodland in central western New South Wales, Australia to extract information on soil, herbage and tree cover. There was a significant correlation (r² = 0-71) between estimated and ground data for tree cover using the TM data, with a mean error of ± 143 per cent, and a mean error of ±11-2 per cent for the bare soil cover estimates. Only general trends were observed using the MSS data. The main areas of confusion were between senesced herbage and soil, and between green grass and the green leaves of trees.