2004
BASE
). An
undle
eters.
/ritten
ly for
ontrol
object
ontrol
hs can
weak
ces "is
were
ym the
global
erence
| their
other
terrain
TOT
P
n
lide: an
fference
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
—> =0.5m/a
Ty,
Figure 6. Movement vectors, showing direction and magnitude
of the displacements between 1953 and 1990(m/a).
source is caused by areas of low reflective contrast, where
digital image matching techniques have difficulty to
automatically identify mass points for DEM extraction. This
may lead to uneven spikes and dips in the model. A more
objective way of accuracy assessment would be to identify a
stable area and compare the photogrammetric derived product
with an independent data set, for example a representative DEM
based on GPS points.
Changes in morphology can be presented by DEMs-of-
difference (change in height), cross-section profiles and
movement vectors from the orthophotographs (change in plane).
Figure 5 shows three-dimensional views created by draping
respectively an orthophotograph and a DEM-of-ditference over
a DEM, which certainly demonstrates the potential of the
approach for visualization.
The horizontal precision is much better, and planar
displacements can be detected from the orthophotographs
(Figure 6). Simple statistical tests revealed the significance of
the horizontal movement vectors. The 14 objects surrounding
the landslide were assumed to be stable positions during all
epochs; the apparent displacements are due to random errors in
the orthophotographs and consistent with the horizontal
precision estimated by the self-calibration adjustment. Two-
sample difference of mean tests proved that the displacements in
x-direction, and total vector length of the 15 selected points on
the landslide are significant greater than these errors, at a 95%
confidence level. Movements in the y-direction are not proven
to be significant. Note that the measured movements over the
period 1971-1973 are not statistical significant, due to the
relative large contribution of noise to the small total
displacement over this short time-interval. The statistical
properties of all samples are summarized in Table 7.
3.6 Comparison with traditional methods
The mean displacement of the landslide over the period from
1953 to 1990 is +0.32m/a (Table 7), varying from +0.11m/a at
the toe up to +0.81m/a in the central. most active part. It is
particularly satisfying that this value is of comparable size to
movement rates found by Ruther et al. (2003), 0.04-0.35m/a
over the last century and up to 0.50m/a in recent years.
4. CONCLUSION AND FURTHER RESEARCH
These initial results demonstrate the value of historical aerial
photographs for quantifying past landslide movements. Due to
the poor base-height ratio of some of the photographs, the
extracted height data were not precise enough to detect
significant vertical displacements. Nevertheless, the accuracy in
plane was sufficient to analyse horizontal movements of the
landslide, even in the cases of a rather low image resolution.
What is encouraging is that the extracted displacements were
consistent with movement rates observed by other researchers,
obtained by direct ground measurement.
Incorporation of more epochs would improve the temporal
resolution; extension of the time series with more recent images
would be particularly useful, as this would provide a overlap
with recent traditional surveys. More detailed analysis of the
displacements in both vertical and planar directions is expected
to gain more insight in the mechanics of the landslide.
In further research a relationship between the extracted mass
movements and climate variables will be determined. The main
challenge is expected to be the temporal scarcity of photo
information. The time between photo epochs will be in the
order of 10 years, while major slide movements take place every
4-5 years on average (Waltham & Dixon, 2000). The ultimate
aim will be the ability to obtain a climate-landslide model that is
able to predict the response to future climate change. The
described method should be applicable to any active landslide
worldwide, provided that historical aerial photography is
available.
1953-1971 1971-1973 1973-1990 1953-1990
Stable objects
sample size 14 13 11 12
x (mean xstd, in n/a) 0.03 +£0.02 -0.13 £0.06 -0.02 +0.02 0.00 £0.01
y (mean +std, in m/a) -0.03 +0.09 0.15 £0.33 -0.02 £0.03 -0.02 £0.04
total (mean xstd, in n/a) 0.09 20.04 0.37 40.17 0.03 20.03 0.04 20.02
On landslide
sample size 15 15 12 12
x (mean xstd, in nva) 0.24 40.17 0.22 x0.41 0.29 +0.21 0.27 +0.18
y (mean +std, in m/a) -0.02 + 0:23 -0.13 £0.72 0.04 £0.17 0.01 30.18
total (mean xstd, in m/a) 0.33 10.17 0.71 £0.47 0.33 +022 0.32 +0.19
Table 7. Statistical properties of the displacement vectors (averaged movements in m/a).
479
