landslides (reference), providing an accuracy of 55 percent
for the algorithm. Additionally, twenty of the misclassified
mapped areas were along rivers and creeks crossing the trans-
portation network, which does not include areas along the
Muskingum riverbank, thus, accounting for 10 percent of the
mapped areas. The reason for these areas being consistently
mapped can be attributed to the amount of erosion generated,
in turn, creating high surface roughness. Nonetheless, the al-
gorithm was able to identify 67 out of 80 mapped landslides in
the inventory map, illustrating that 84 percent of the mapped
landslides from the reference were identified. Although some
of the mapped areas did not overlap the reference map, they
were adjacent to these areas (see Figure 7C). Further analysis
is necessary to verify that these mapped areas are indeed not
new developing landslides or existing landslides that have
developed further. Moreover, additional analysis is required to
evaluate why some of the inventory mapped landslides were
overlooked by the proposed algorithm. One reason for over-
looking mapped landslides (reference) is the amount of sur-
face roughness exhibited within the landslides (see SW corner
of Figure 7A, West of road for Figure 7B and Figure 7D). The
amount of surface roughness is not sufficient to delineate them
from stable terrain. Therefore, these mapped landslides (refer-
ence) will go undetected, until enough surface roughness is
displayed from experienced mass movement.
Conclusions
Landslide susceptibility mapping using remote-sensing
techniques may never completely replace traditional map-
ping methods of field inspection, aerial photograph inter-
pretation, and contour map analysis. Moreover, the mapping
methods presented in this and other studies often rely on
objective topographic data that relies on the morphologic
expressions in the area studied, and often cannot differentiate
between adjacent landslides. However, as the spatial resolu-
tion, accuracy, and availability of remote-sensing technology
increases, new landslide susceptibility mapping methods will
provide efficient tools that can assist traditional methods. The
proposed approach quantifies and identifies landslide surface
features producing results that can potentially become useful
in the prevention of future hazardous events.
Although, the generation of landslide maps remains a sub-
jective and time consuming task, airborne lidar provides new
opportunities for mapping the topographic features found in
small landslides. lidar technology has become both more ac-
cessible and affordable, and the advancements in airborne lidar
have allowed for a new method to map landforms, including
landslides, over broad swaths of terrain at higher spatial resolu-
tions and accuracy. To our best knowledge, the literature has not
capitalized on airborne lidar-derived
DEM
s to investigate small
landslide susceptibility mapping at sub-meter scales over large
Figure 6. Topographic variability of four segments along SR-666. The surface feature used to depict the topographic variability was
roughness. The higher variability the rougher the surface, which is darker in the figure.
Figure 7. Comparison between the mapped landslide suspect areas by the proposed technique and those provided in the reference
inventory map. The mapped landslides with a solid black line are those mapped by the proposed algorithm, and those mapped by the
dashed line in white are the landslides mapped in the inventory map. The underlying DEM is a rasterized hillshade DEM.
246
March 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
