Combined refinement for DEM using low-pass filters and a fitting model

摘要

In this study, we test the impact of using low-pass linear filters on the vertical accuracy of SRTM (Shuttle Radar Topography Mission) digital elevation model (DEM). We adopt three weight kernels to utilise low-pass filtering methods such as Distance Weighting (DW), Gaussian (G) and Inverse Distance (ID). The levelling data are used to validate the impact of the filtering methods on the DEM by adopting different values for the parameters of the weight kernels. As a result, we find that the DW and ID are more consistent with respect to the specified kernel power rather than G filter which shows a big range of bias variation. Another test is performed by generating diagonal transects over the filtered and unfiltered (SRTM) data, 90% of filtered datasets are consistent with SRTM data in a range of ±0.5 and 77% G data for the same range. Finally, the bias due to systematic errors is removed using stochastic 5-Parameter model. This shows an improvement ~1 m in the vertical accuracy with a standard deviation of about 0.8m.