降雨誘發淺層山崩模型土壤強 度參數逆分析之比較與驗證 Adviser:董家鈞、劉家男 Student:陳麒任 • Introduction – Objective – Literature Review • Methodology – Data base – Back analysis • Result and Discussion • Conclusions and Recommendation Qualitative analysis Empirical method Quantitative analysis Statistic method Discriminant.
Download ReportTranscript 降雨誘發淺層山崩模型土壤強 度參數逆分析之比較與驗證 Adviser:董家鈞、劉家男 Student:陳麒任 • Introduction – Objective – Literature Review • Methodology – Data base – Back analysis • Result and Discussion • Conclusions and Recommendation Qualitative analysis Empirical method Quantitative analysis Statistic method Discriminant.
Slide 1
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 2
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 3
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 4
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 5
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 6
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 7
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 8
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 9
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 10
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 11
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 12
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 13
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 14
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 15
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 16
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 17
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 18
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 19
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 20
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 21
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 22
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 23
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 24
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 25
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 26
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 27
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 28
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 29
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 2
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 3
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 4
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 5
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 6
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 7
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 8
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 9
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 10
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 11
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 12
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 13
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 14
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 15
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 16
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 17
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 18
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 19
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 20
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 21
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 22
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 23
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 24
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 25
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 26
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 27
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 28
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion
Slide 29
降雨誘發淺層山崩模型土壤強
度參數逆分析之比較與驗證
Adviser:董家鈞、劉家男
Student:陳麒任
• Introduction
– Objective
– Literature Review
• Methodology
– Data base
– Back analysis
• Result and Discussion
• Conclusions and Recommendation
Qualitative analysis
Empirical method
Quantitative analysis
Statistic method
Discriminant analysis
Logistic regression
Conditional Probability Approach
Artificial intelligence
Fuzzy Theory
neural network
Deterministic analysis
Rainfall trigger landslide
Earthquake trigger landslide
Cohesion
Friction
angle
Lab test
Lab test
hydraulic
hydraulic
conductivity
conductivity
unit weight
of soil
In situ test or
In situ test or
depth
Empirical methods EmpiricalSoil
methods
Remote sensing
Observed landslide
inventory
DEM
Remote sensing
Slope
unit weight
of soil
Deterministic
analysis
Deterministic
analysis
Parameters
Soil depth
DEM
Slope
Rainfall intensity Rainfall intensity
Predicted landslide
inventory
Godt et al. (2008)
4
Observed
Predicted
1) Extensive work to get reliable data. [林衍丞,2009]
2) Strength parameter and hydraulic parameter are difficult to
obtain. [李錫堤,2009]
3) There are scale issues involved in the translation of laboratory
values to the field problem. [Guimaraes, 2003]
4) Back analysis of strength has advantages over lab testing in
that the scale is much larger. [Gilbert ,1998]
5) Back analysis is reliable only when the model and all
assumptions are reasonable and accurate representations of
the real system[Deschamps, 2006]
• Exist many back analysis criterion.
Efficiency: (
+
)/(
Sensitivity:
Specificity:
Observed
Predicted
+
/(
/(
+
+
+
+
)
)
Observed
Unstable
Stable
Unstable
True Positives
(TP)
False Positives
(FP)
Stable
False Negatives
(FN)
True Negatives
(TN)
Predicted
7
)
However, the output of back analysis is usually uncertain because
of the random factors existing in the problem. [Zheng, 2008]
Methodologies used for back analysis can be classified into two
groups, i.e., deterministic method and probabilistic
method.[Zhang, 2010]
1) Compare the existing back analysis criterion.
2) Compare the result of deterministic method and
probabilistic method.
Rainfall-induced landslide model
This research use TRIGRS, a Fortran program developed by USGS.
The Transient Rainfall Infiltration and Grid-Based Regional SlopeStability.
Theoretical Basis
• Infinite-slope stability
– Landslide with planar failure surfaces.
– Slide depth is much smaller than length and width.
where c’ is soil cohesion for effective stress, Φ’ is the soil friction angle for effective
stress, γw is unit weight of groundwater, and γs is soil unit weight, β is slope angle, ψ is
pressure head.
11
Back analysis parameters
林衍丞,2009
Collect the back analysis criterion
1) Maximum Efficiency(林衍丞,2009) 。
2) Maximum AUC (林衍丞,2009) 。
3) Efficiency greater than 80%, Sensitivity
greater than 60% and Specificity greater
than 90%(中興工程顧問社 ,2004)。
FS=1.5
FS=1
ROC
FS=0.5
Sensitivity
4) Maximum Develop Sensitivity
Specificity
林衍承(2009)
Study Area
14
Input Data
Soil depth
15
Input Data
Slope
Storm event
2001/7/29 ~ 2001/7/30
Input parameters
Unit weight
parameter
of soil
(kN/m3)
Value
20
saturated
vertical
hydraulic
conductivity
(m/s)
5E-5
Initial
infiltration
rate (m/s)
1E-6
initial
hydraulic
water-table
diffusivity (m2/s)
depth (m)
5E-5
Equal to soil
depth
Consider Salciarini(2008) , Godt(2008)
Result and Discussion
(A) Develop sensitivity
(B) Efficiency
(C) Efficiency greater than 80%,
Sensitivity greater than 60%
and Specificity greater than
90%
(D) AUC
Criterion B :Efficiency
1.
2.
3.
4.
Low failure ratio
Overestimate parameters
Underestimate landslide
Select parameters hardly
Sensitivity Efficiency Specificity
0.076
0.890
0.994
Develop
sensitivity
0.072
Criterion A,C
1.
2.
3.
4.
Good constrain
Low friction angle
High cohesion
Assumption problem
(depth, variable)
Sensitivity Efficiency Specificity
0.717
0.804
0.816
Develop
sensitivity
0.293
Author
Date
Sensitivity Specificity Efficiency
Godt
2008
0.42
0.84
0.80
Tiwari (2000,2005) assumed factor of safety is equal to 0.98 for
back analysis cohesion and friction angle.
Sensitivity= 0.4~0.44
Specificity=0.80~0.88
Efficiency=0.75~0.85
Sensitivity
0.4050
Efficiency Specificity
0.8258
0.8794
Develop
sensitivity
0.2080
Bayesian theorem:
Updates a probability given new information
雙
變
量
常
態
分
布
(mean)
凝聚力
山崩
摩擦角
不山崩
(coefficient of
variance)
cohesion
8 kPa
0.2
Friction angle
30°
0.1
Chen et al.(2005)
Zhang et al.(2010):Back analysis of slope failure with Markov chain Monte Carlo simulation
Gilbert et al.(1998):Uncertainty in back analysis of slopes: Kettlemen Hills case history
P
P
1
Fs
0.42
0.84
0.8
多
變
Sensitivity 量
常
Specificity 態
Efficiency 分
布
Mean of
Mean of cohesion
COV of
COV of
friction
(kPa)
cohesion
friction angle
angle(°)
After update
Before update
1600
8.87
0.13
8
0.2
(a)
0.03
30
0.1
800
(b)
600
Number of sample
1200
Number of sample
35.58
800
400
400
200
0
0
30
31
32
33
34
35
36
37
Friction angle
38
39
40
41
3
4
5
6
7
8
9
10 11 12 13 14 15
Effective soil cohesion