DOUBLE GAUSSIAN FILTERING TO SUPPRESS NOISE AND IMPROVE IDENTIFICATION OF NEW LANDSLIDES ON DEM DIFFERENCE MAPS

Simple subtraction of before-and-after digital elevation models (DEMs) created using LiDAR surveys, multibeam echosounder (MBES) surveys, or other remote sensing technologies such as structure from motion (SfM) photogrammetry can provide evidence of landslide occurrence over time. However, DEM difference maps covering steep forested slopes prone to landslides can be noisy and difficult to interpret when the elevation changes associated with downslope movement are near the magnitude of the noise. This is in part because subtraction of one noisy DEM from another compounds the noise. Using as an example several small landslides that occurred in colluvium derived from the Ordovician Kope Formation— known for chronic slope stability problems along the Ohio River valley in the Cincinnati and northern Kentucky area—between 2007 and 2012 airborne LiDAR surveys, we show how a two-step Gaussian filtering approach can help to suppress noise while retaining the characteristic landslide signature. The first step is to remove nonstationary bias from the DEM difference map by applying a Gaussian filter and subtracting the filtered difference map from the original difference map, creating a residual map with near zero local mean values. The next step is to apply a second Gaussian filter to the residual map to remove noise while preserving the landslide signature. In both cases, the choice of the Gaussian filter moving window size is subjective but guided by geologic insight. By way of comparison, we show that the double-Gaussian approach can produce maps that are cleaner and more easily interpretable (especially for automated mapping applications) than those produced by simple subtraction or purely statistically threshold approaches. Such information is useful for optimizing the timing of resurveys as well as understanding the limits of detection from pairs of surveys.