Assessment of digital elevation model accuracy using ALOS-PRISM stereo imagery

Triplet-stereo imagery from the PRISM instrument on the ALOS satellite has been processed to a digital elevation model (DEM) for use in soil prediction models. A number of artefacts in the ALOS-PRISM imagery appear to limit the quality of the stereo imagery, and thus the derived DEM. Even/odd column striping has been reduced in its effect by a change in processing in early 2007, but some residual striping remains. Secondary striping suppression reduces this striping effect to some extent. JPEG compression, which was severe in imagery processed before early 2007, has been suppressed to some extent by post-processing. Some improvements are seen between 8 × 8 pixel blocks in the image, but distortion within them is still evident in imagery processed after early 2007. Estimates of the standard deviation of the DEM error are strongly affected by the quality of the control points, and the study suggests that more points with higher accuracy are required. A Manawatu DEM generated from ALOS-PRISM imagery has a mean of -4.64 m (95% confidence interval [-6.87, -2.41]m) and a standard deviation of 6.48m (95% equal-tail confidence interval [4.46, 11.8]m). The DEM standard deviation estimate has relatively broad confidence limits as a result of the small number of geodetic marks (10) in the test area, and is likely to be affected by the quality of the control points, which were derived from LINZ interpolated contours.