(
ODOT
) Office of Geotechnical Engineering, and provided im-
portant information about the general locations of landslides
affecting the road prism. An updated landslide inventory map
was compiled by a team of experts from Kent State University
and The Ohio State University through contour map analysis,
geo-hazard inventory evaluation and on-site validation in the
summer of 2012. The updated landslide inventory was used
for the investigation. Typical landslides affecting the road
prism are: rotational, translational, complex, rockfall debris,
and mudslides. The slopes for areas of instability range from
18° to 80°, in which the most frequently slope observed was
45°; additionally, the landslides described have a range of
ages per the historical documents. The importance of the
inventory map is that it provides a reference against which
we can evaluate the performance of our proposed approach. A
limitation found in the reference map is that it only provides
the extents of the mapped landslides and does not offer ad-
ditional information about the rate of change experienced.
The mapped landslides provided in the inventory range from
200 m
2
to 27,000 m
2
in area. The soil map used for this study
was available online from the Ohio Department of Natural
Resources (
ODNR
) website
-
id/9051/Default.aspx
).
Methodology
The effects of mass movement are important and greatly
dependent on their spatial pattern of occurrence, frequency,
and amount of activity (McKean and Roering, 2004). The
temporal processes of landslides can reveal a wealth of
information regarding the magnitude of surface deforma-
tion experienced and the expected change over time. While
temporal changes cannot be revealed from individual surface
models, identifying landslide-specific spatial features from
single surface models is important, as not all the changes
detected by temporal analysis represent landslide suspect
areas. This study is focused on examining and evaluating
single surface models, and the developed method can serve as
a tool to filter landslide suspect areas. Landslides are known
to have rougher surfaces than neighboring stable terrain. This
is due to the mechanics, subsidence, and surface deforma-
tion experienced. The surface roughness of landslide (bottom)
terrain experiences higher topographic variability than stable
(top) terrain as illustrated in Figure 2. McKean and Roering
(2004) and Glenn
et al.
(2006) exploited the surface roughness
to detect and map landslides, and confirmed that the surfaces
of landslides are rougher than neighboring stable terrain. For
these reasons, the surface roughness will be the focus of the
proposed algorithm.
The objective of the approach is to identify surface features
indicative of landslide activity and map their locations in the
study area. The process to identify landslide surface features
is as follows: (a) filter airborne lidar point cloud to contain
bare-earth points only, (b) rasterize the bare-earth point cloud
using Kriging interpolation method, (c) perform surface
feature extraction, (d) classify lidar-derived
DEM
, (e) perform
post-classification filtering, and lastly (f) map areas experienc-
ing landslide activity. The feature extraction algorithms used
are described in the following sections.
Feature Extraction
To extract and quantify the amount of surface roughness
observed in the terrain, the following eight geomorphological
features were utilized: aspect, hillshade, roughness, slope,
eigenvalue ratios (
λ
1
/
λ
2
and
λ
1
/
λ
3
), customized Sobel operator,
and the resultant length of orientation vectors. The selected
feature extraction methods are further discussed below. Some
of the surface feature extraction methods selected have been
used to expose various topographic patterns (e.g., McKean
and Roering, 2004; Glenn
et al.
, 2006) and were therefore
prime considerations. The standard algorithms available in
the
MATLAB
TopoToolbox by Schwanghart and Kuhn (2010)
were used for the evaluation of aspect, hillshade, roughness,
and slope. Fixed sampling windows of size (9 × 9) were used
to evaluate the direction cosine eigenvalue ratios and length
of orientation vectors. Furthermore, a statistic measure of the
standard deviation is evaluated from small sampling win-
dows of a fixed size (9 × 9
)
to define the local topographic
variability of aspect, hillshade, roughness, slope, resultant
length of orientation vectors, and customized Sobel opera-
tor. Areas experiencing higher degrees of surface deformation
will illustrate higher topographic variability, thus, delineating
rough and smooth terrain.
Aspect
Slope orientation is the compass direction a land surface
faces. To evaluate the slope orientation, also known as aspect,
for a
DEM
grid point of a (3×3) local neighborhood,
Z
11
, shown
in Equation 1, the surface normals need to be computed. Sub-
sequently, the mapping system needs to be converted from a
two-dimensional Cartesian coordinate system to a polar
coordinate system:
θ
=
arctan
N
N
x
y
, where,
θ
is the angle in
the polar coordinate system, and
N
x
and
N
y
are the surface
normals in the east-west and north-south direction, respec-
tively. Finally, the slope orientation of a cell can be computed:
ASPECT
=
θ
* 180° + 180°.
Z Z Z
Z Z Z
Z Z Z
02
12
22
01
11
21
00
10
20
(1)
Hillshade
The relief depiction of a grid point in a
DEM
is described by
the lighting effect of the angle between the surface and the
incoming light beam. The approach uses the illumination
Figure 2. The figure illustrates surface normals representing
topographic variability (roughness) in a DEM. Smooth terrain
(top), illustrates less variation. Rough terrain (bottom), illus-
trates higher variability from McKean and Roering, 2004.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
March 2015
241
