Geoscience Reference
In-Depth Information
where:
NL
\
Eij
ð
Þ
:
number of landslide pixels in domain
total number of landslides, or
P
NL
\
E
ij
N(L):
N(Eij): number of pixels in domain
N(A): total number of pixels in domain, or
P
N(E
ij
)
N: proportion of landslide area
D:
proportion of non-landslide area
Equation
8
shows the final result of Bel ratio in each class.
j T
ð
Eij
=
X
Bel ¼
j T
ð
Eij
ð
8
Þ
Similarly, Eqs.
9
and
10
, were used to find the Dis ratio in each class. Finally,
Eq.
11
was used to calculate the Unc and Eq.
12
gave the Pls result.
Tp
j
ð
Þ
Eij ¼½
ð
N
ðÞ
N
ð
L
\
Eij
ÞÞ=
N
ð
L
Þ=
½ N
ðÞ
N
ðÞ
N Ei
ðÞþ
NL
\
Eij
ð
ð
Þ
Þ=ð
N
ðÞ
N
ð ÞÞ
¼ K
=
H
ð
9
Þ
where:
K: proportion of landslides that do not occur.
H: proportion of non-landslide areas in other attributes outside class
Þ
Eij
=
X
j Tp
ð
Tp
Þ
Eij
Dis ¼
ð
j
ð
10
Þ
Unc ¼ 1
Dis
Bel
ð
11
Þ
Pls
¼
1
Dis
ð
12
Þ
The EBF model gave the spatial distribution of landslide prediction map
through different zones showing the degree of uncertainty of the same zone (Park
2010
). Therefore, some conclusion can be drawn in terms of model capability:
(a) Degrees of belief: showed the susceptible areas.
(b) Degrees of disbelief: showed the non-susceptible areas.
(c) Degrees of uncertainty: showed where the evidences are insufficient to provide
the proofs for landslide information, or guide for further field assessment.
(d) Degrees of plausibility: represented all the integrated maps evidence except
the disbelief map. Generally it shows where spatial evidences are sufficient. Or
evidences are inefficient to prove where the landslide triggered factor will
effect.
Moreover, data-driven EBF model could compromise the results if spurious
evidence arises from insignificant spatial associations between landslides and
conditioning factors such as the case of geology. Additionally, the model can
produce acceptable evidences even with large number of input dataset. Finally,
two LSMs were produced, one using the cluster selection technique and the other
