Digital filtering of generic topographic data in geomorphological research

Abstract

High resolution terrain models generated from widely available Interferometric Synthetic Aperture Radar (IfSAR) and digital photogrammetry are an exciting resource for geomorphological research. However, these data contain error, necessitating pre-processing to improve their quality. We evaluate the ability of digital filters to improve topographic representation, using: (1) a Gaussian noise removal filter; (2) the proprietary filters commonly applied to these datasets; and (3) a terrain sensitive filter, similar to those applied to laser altimetry data. Topographic representation is assessed in terms of both absolute accuracy measured with reference to independent check data and derived geomorphological variables (slope, upslope contributing area, topographic index and landslide failure probability) from a steepland catchment in Northern England. Results suggest that proprietary filters often degrade or fail to improve precision. A combination of terrain sensitive and Gaussian filters performs best for both datasets, improving the precision of photogrammetry digital elevation models (DEMs) by more than 50 per cent relative to the unfiltered data. High frequency noise and high magnitude gross errors corrupt geomorphic variables derived from unfiltered photogrammetry DEMs. However, a terrain sensitive filter effectively removes gross errors and noise is minimised using a Gaussian filter. These improvements propagate through derived variables in a landslide prediction model, to reduce the area of predicted instability by up to 29 per cent of the study area. IfSAR is susceptible to removal of topographic detail by over-smoothing and its errors are less sensitive to filtering (maximum improvement in precision of 5 per cent relative to the raw data).