High-accuracy terrain modelling for soil mapping using ALOS-PRISM imagery

A digital elevation model (DEM) has been generated from ALOS-PRISM triplet stereo imagery of the Manawatu and Wanganui regions, New Zealand, for use in high-country soil mapping. The PRISM imagery suffers from block and high-frequency JPEG compression artifacts, as well as column and inter-CCD striping. Although some of these artifacts have been suppressed in the PRISM processing chain from October 2007, the residual effect of these sensor problems is to generate poor outlier estimates of stereo disparity; this increases the manual post-processing effort needed. Column striping was suppressed by equalizing between column mean differences; anomalous column striping was subsequently suppressed by equalizing between the column mean and the median column mean; JPEG high-frequency artifacts were ameliorated by using a repeated-reconstruction approach separately for even and odd columns, using estimates of the JPEG compression table. Using geodetic control points for validation, the model standard deviation is estimated to be, at best, 2.9m.