Comparative analysis of surface roughness algorithms for the identification of active landslides

Parameters correlated to surface roughness are quite commonly used to describe landslide activity in quantitative
geomorphology. Previous studies proved that topographic roughness is closely related to both landslide mechanics
and features. A number of different techniques have emerged over the years to describe quantitatively
the great variety of landforms and processes that affect unstable slopes. In this work we perform a comparative
analysis of severalmethods used in literature to compute surface roughness (root mean square applied to elevation
and slope grids, eigenvalue ratios, semivariance, discrete Fourier transform, continuous wavelet transform
and wavelet lifting scheme) in order to evaluate quantitatively which algorithms are best suited to discriminate
active landslides and to predict them for automated mapping purposes. A first test was carried out on artificial
surfaces simulating different roughness patterns encountered in nature, so to highlight advantages and limits
in controlled conditions. Then, the algorithms were applied to LiDAR datasets of two earth flow case studies in
the Northern Apennines, Italy.
Results obtained by using “effect-size” statistical test to objectively quantify the capability of the different algorithms
of discriminating active landslide slopes from other slope types showed that most algorithms perform
reasonablywell and that simple techniques (RMS-based and wavelet lifting scheme) achieve equal or sometimes
even better results thatmore complex ones. Results fromthe use of roughness indexes for the prediction of landslide
slopes in automated mapping showed that non-forested active slopes could be predicted bymostmethods
with an accuracy greater than 85% and that most methods had a 15% drop in prediction accuracy in forested
active slopes. Results also proved that increasing the size of the moving window has minor beneficial effects in
predictive capability, suggesting that small size of pixels and moving windows should be used to retain a full
resolution of surface conditions in slopes.