SPATIAL MODELS FOR THE DEFINITION OF LANDSLIDE

SPATIAL MODELS FOR THE DEFINITION
OF LANDSLIDE SUSCEPTIBILITY AND
LANDSLIDE HAZARD
J.L. Zêzere
Centre of Geographical Studies
University of Lisbon
FORM-OSE POST-GRADUATE TRAINING SCHOOL
Living with hydro-geomorphological risks: from theory to practice
14-19 September 2004, Strasbourg - France
CONCEPTUAL MODEL OF LANDSLIDE RISK
Dangerous Phenomena
Vulnerable Elements
Rockfall
Topple
Slide
Spread
Flow
….
Population
Buildings
Infrastructures
Economic activities
Cultural and environmental values
….
Landslide hazard
Vulnerability
LANDSLIDE RISK
adapted from Panizza (1990 )
Landslide Hazard = probability of occurrence of a
potentially damaging phenomenon [landslide]
within a given area and in a given period of time.
1. BASIC CONCEPTS
Landslide Hazard:
“the probability of occurrence of a potentially
damaging phenomenon [landslide] within a
given area and in a given period of time.”
CRUCIAL ELEMENTS IN THE PREDICTION OF
FUTURE LANDSLIDE BEHAVIOUR:
1) SPATIAL LOCATION (“WHERE”?)
2) TIME RECURRENCE (“WHEN”?)
(Varnes et al., 1984)
Rt = (E) (Rs) = (E) (H x V)
3) INTENSITY / MAGNITUDE (“HOW POWERFUL”?)
TYPICAL METHODS USED TO DEFINE LANDSLIDE
PRONE ZONES AT A REGIONAL SCALE:
H – Hazard
V – Vulnerability (degree of loss; 0-1)
E – Vulnerable Elements (Value of...)
Rs – Specific Risk (H x V)
Rt – Total Risk
1)DIRECT APPROACH (Geomorphological)
2) INDIRECT APPROACHES (Quantitative and
Semiquantitative)
¾ Knowledge-based (index)
Applied over
¾ Statistical (data-driven)
predefined
terrain units
¾ Deterministic
1
BASIC ASSUMPTION OF BOTH DIRECT AND INDIRECT
APPROACHES:
FUTURE LANDSLIDES ARE MORE LIKELY TO OCCUR
UNDER THE SAME GEOLOGICAL AND
GEOMORPHOLOGICAL CONDITIONS THAT LED TO PAST
SLOPE INSTABILITY.
SUSCEPTIBILITY VS HAZARD
Most regional landslide ‘hazard’ assessments (both direct
and indirect) provide a ranking of terrain units only in terms
of susceptibility (“spatial probability”), not including the
temporal component of the hazard.
“PAST AND PRESENT ARE KEYS TO THE FUTURE”
KEYS FOR PREDICTION OF FUTURE LANDSLIDES:
DIRECT APPROACH
Gabbione, Pavia
Italy
Rossetti (1997)
¾ MAPPING PAST AND RECENT SLOPE MOVEMENTS
¾ IDENTIFICATION AND MAPPING OF THE
CONDITIONING OR PREPARATORY FACTORS OF
SLOPE INSTABILITY
ASSESSMENT OF LANDSLIDE SUSCEPTIBILITY
Direct Landslide Susceptibility Assessment (Calhandriz Area)
LANDSLIDE SUSCEPTIBILITY
Expression of the likelihood that a landslide will occur in an
area based on the local terrain conditions, not including the
return period or the probability of occurrence of the instability
processes.
2
Indirect Landslide Susceptibility Assessment (Information Value Method)
FROM LANDSLIDE SUSCEPTIBILITY
TO LANDSLIDE HAZARD
By definition, a hazard map should include an evaluation of the
probability of occurrence of new landslides, thus implying the
consideration of a time dimension.
In the case of rainfall induced landslides the statistical
analysis of rainfall data may enable both the definition
of the triggering threshold and calculation of the
recurrence interval.
UNCERTAINTIES AND DRAWBACKS
1) DATA LIMITATIONS
¾ UNCERTAINTIES IN LANDSLIDE IDENTIFICATION
AND MAPPING
¾ QUALITY, QUANTITY AND RELEVANCE OF THE
AVAILABLE INFORMATION:
9 The discontinuous nature in space and time of slope
failures
9 The lack of a complete historical data concerning
the frequency of landslides
9 The difficulty of identifying the causes, the triggering
factors and cause-effect relationship
2) MODEL SHORTCOMINGS
¾ EFFECTIVENESS AND RELIABILITY OF THE
AVAILABLE MODELS
¾ DIFFICULTY IN EXTRAPOLATION OF LOCAL DATA
TO LARGER AREAS
ITC, ILWIS Manual
3
Translational movements susceptibility assessment (Information Value Method)
3) ADDITIONAL PROBLEM
¾ THE DIFFERENT SPATIAL INCIDENCE OF
DIFFERENT TYPES OF SLOPE MOVEMENTS,
NORMALLY RELATED TO DISTINCT THRESHOLDS
CONDITIONS CONCERNING PREPARATORY AND
TRIGGERING FACTORS.
ASSESSMENT OF LANDSLIDE HAZARD
FOR EACH TYPE OF SLOPE MOVEMENTS
Indirect Landslide Susceptibility Assessment (Information Value Method)
Shallow translational slides susceptibility assessment (Information Value Method)
Rotational movements susceptibility assessment (Information Value Method)
VALIDATION OF SUSCEPTIBILITY
AND HAZARD
WHY TO VALIDATE ?
Evaluation of the predictive power of models with
respect to future slope movements
Strictly speaking, validation of the prediction of future
landslides is only possible with the ‘wait and see’
procedure.
4
24
0
Very high
II
High
III
Moderate
ALARM PROJECT
FANHÕES-TRANCÃO
TEST SITE
Spain
Lisbon
28
0
I
Study
area
Ta
g
Susceptibility classes:
EDP
REGIONAL FRAMEWORK
Oporto
r
0
20
0
24
Landslide susceptibility
assessment in the Trancão
river valley (Direct approach,
1988)
us
riv
e
16
0
12
0
160
240
200
0
12
Atlantic Ocean
280
IV
200
Low
1
240
160
CREL motorway (1995)
December 1995 / January 1996
landslides
2
• Monocline structure dipping 5° to 25° to S and SE.
• Heterogeneous bedrock (limestones, sandstones, basalts, volcanic
tuffs, marls, clays) dating from upper Jurassic to Miocene.
• Cuestas, strongly dissected by fluvial cutting.
• Maximum altitude = 350m; steep slopes on catacline valleys.
200
0
120
80
200 m
160
Fanhões valley
Trancão valley
VALIDATION OF SUSCEPTIBILITY
AND HAZARD
VALIDATION FOR WHAT ?
Evaluation of the predictive power of models with
respect to future slope movements
Strictly speaking, validation of the prediction of future
landslides is only possible with the ‘wait and see’
procedure.
Proposed method: spatial/ time partitioning of the
spatial landslide databases.
2. LANDSLIDE SUSCEPTIBILITY, HAZARD
ASSESSMENT AND ZONATION
ge
olo
gy
Supporting patterns
selected by expert
Prediction map:
resource potential,
hazard or impact
slo
pe
Expert
General
methodology
from data
capture and
treatment to
landslide
susceptibility
assessment and
validation
DATA CAPTURE AND DATA
TREATMENT
Contour
lines
(5 m)
Altitude
points
Verification
Rectification
DEM
(pixel: 5m)
Documentation
Air-photo interpretation
algorithms
Field work
Slope aspect
(continuous)
Slope angle
(continuous)
CARTHOGRAPHIC DATABASE
Landslide
map
Superficial
deposits
Lithology
Land use
Geomorph.
units
Slope
profile
Slope
angle
Slope
aspect
Independent data layers (categorical)
res know
ou
rce n ha
s o zard
r im s,
pa
cts
Lan
du
se
DATA INTEGRATION
Lithology
Validation
supporting pattern
landslides
Superficial
deposits
landslides
landslides
Partition
(temporal criteria)
Mathematician
Validation group
[age > 1979]
(54 cases)
Geomorph.
units
Estimation group
[age <= 1979]
(46 cases)
Land use
landslides
Slope
profile
landslides
Slope
angle
Slope
aspect
landslides
landslides
LANDSLIDE
SUSCEPTIBILITY MAP
Susceptibility
prediction
image
Data
integration
CLASSIFICATION
INTERPRETATION
PREDICTION-RATE
CURVE
Assessment of Landslide Risk and Mitigation in Mountain Areas
EVG1-2001-00018
SUSCEPTIBILITY ASSESSMENT AND VALIDATION
5
CONSTRUCTION OF A CARTHOGRAPHIC DATABASE
LITHOLOGY
DATA CAPTURE AND DATA
TREATMENT
Altitude
points
Verification
Rectification
Contour
lines
(5 m)
DEM
(pixel: 5m)
Documentation
Air-photo interpretation
algorithms
Field work
Slope angle
(continuous)
Slope aspect
(continuous)
CARTHOGRAPHIC DATABASE
Landslide
map
Lithology
Superficial
deposits
Geomorph.
units
Land use
Slope
profile
Slope
angle
Slope
aspect
Independent data layers (categorical)
SLOPE ANGLE
SUPERFICIAL DEPOSITS
(degrees)
SLOPE ASPECT
GEOMORPHOLOGICAL UNITS
Flat areas
N
NE
E
SE
S
SW
W
NW
6
Landslide morphometric parameters of the
FanhõesFanhões-Trancão test site
LAND USE /
VEGETATION COVER
Landslide
types
Rotational
slides
Translational
slides
Shallow transl.
slides
Total
TRANSVERSAL SLOPE
PROFILE
Mean
area
(m2)
6,544
Total
area
(m2)
137,415
Mean
volume
(m3)
14,650
Total
volume
(m3)
307,653
6,699
174,185
21
14.3
Mean
depth
(m)
5.3
26
17.7
3.4
6,429
167,151
100
68.0
1.0
1,422
142,176
357
35,357
147
100.0
2.1
3,039
446,742
3,542
517,195
N
(%)
SUSCEPTIBILITY ASSESSMENT
General assumption:
Future landslides can be predicted by statistical relationships
between past landslides and the spatial data set of the
conditioning factors (e.g. slope, aspect, slope profile,
geomorphology, lithology, superficial deposits, land use, etc.).
DATA INTEGRATION
LANDSLIDES
DATA INTEGRATION
Lithology
landslides
Superficial
deposits
landslides
Geomorph.
units
landslides
Land use
landslides
Slope
profile
landslides
Slope
angle
Slope
aspect
landslides
landslides
Ri ver
Trancão
ões
Fanh
Riv er
Rotational slides
Translational slides
Shallow translational slides
7
Shallow translational slides
Fanhões-Trancão test site
DATA INTEGRATION (BAYESIAN PROBABILITY)
Joint Conditional Probability
Function
Non-classified
susceptibility map
(based on the complete
landslide data set -100
cases).
0.0767
DATA INTEGRATION (BAYESIAN PROBABILITY)
Success-rate curve of the susceptibility assessment
based on the complete landslide data set
Prior probability of finding a landslide
affected area/total area
Prior probability of finding a class of a layer
class area/total area
Conditional probability of finding a landslide in each class,
for each layer
⎞
⎛
1
⎟⎟
1 − ⎜⎜1 −
class
area
⎠
⎝
affected area in the class
DATA INTEGRATION (BAYESIAN PROBABILITY)
Probability of finding a landslide, given n data layers, using
the conditional probability integration rule (Chung & Fabbri,
1999):
( Pp L1 × Pp L2 ×...× Pp Ln )(Cp L1 × Cp L2 × ⋅ ⋅ ⋅ × Cp Ln )
L
Ppslide n−1 × ( L1 × L2 × ⋅ ⋅ ⋅ × Ln )
where L1, L2, ..., Ln are the several data layers used as independent variables, (L1×L2×
... ×Ln) represents the prior probability of finding the n data layers in the test site, Cp
is the conditional probability of finding a landslide in a class of each layer, and Pp and
LANDSLIDE SUSCEPTIBILITY VALIDATION
AND CLASSIFICATION
Landslides
Data set
Partition
(temporal criteria)
Validation group
[age > 1979]
(54 cases)
Estimation group
[age <= 1979]
(46 cases)
LANDSLIDE
SUSCEPTIBILITY MAP
Susceptibility
prediction
image
Data
integration
CLASSIFICATION
INTERPRETATION
PREDICTION-RATE
CURVE
SUSCEPTIBILITY ASSESSMENT AND VALIDATION
Ppslide are the prior probabilities of finding, respectively, a class and a landslide in
the study area.
8
Shallow translational slides
Fanhões-Trancão test site
Joint Conditional Probability
Function
Non-classified
susceptibility map
TRIGGERING ASSESSMENT
Landslides in the study area have a clear climatic signal.
S.Julião do Tojal Annual precipitation
Slope instability events
1400
(based on Estimation Group
landslides - age ≤ 1979, 46
cases)
1200
R (mm)
1000
MAP
800
600
400
0.0634
200
0
Archive
investigation
Prediction-rate curve of the susceptibility assessment based on Estimation
Group landslides (age ≤ 1979, 46 cases) and compared with Validation Group
landslides (age > 1979, 54 cases).
Field
work
Interviews
Rainfall analysis
(daily data)
Pe rc e n ta g e o f p re d ic te d
v a lid a tio n g ro u p la n d s lid e s
Prediction-Rate Curve
100
90
80
70
60
50
40
30
20
10
0
IV
III
II
V
Reconstruction of antecedent rainfall
from 1 to 90 days
(Pxn = P1 + P2 +…Pn)
Reconstruction of past
landslide activity dates
84%; 100%
Return
periods
(Gumbel
law)
35%; 86%
18%; 70%
I
0
8%; 41%
10
20
Landslide
type A
30
40
50
60
70
80
90
Types of
landslides
100
Percentage of study area predicted as susceptible using
estimation group landslides
Landslide
type …..
Shallow translational slides
Fanhões-Trancão test site
Joint Conditional Probability
Function
Susceptibility map
classified according to
the prediction-rate
curve.
I – Top 8%
II – 8-18%
III – 18-35%
IV – 35-84%
V – 84-100%
Methodology for rainfall triggering
of landslides analysis
Temporal occurrence of rainfall triggered landslides in Lisbon area
Episode
Date (yy/mm/dd)
1
1958/12/19
Critical rainfall
amount/duration
mm (dd)
149 (10)
2
1959/03/09
175 (10)
4
a
3
1967/11/25
137 (1)
60
a,d
4
1968/11/15
350 (30)
6.5
b
Return period
(years)
Landslide
typology
2.5
a
5
1978/03/04
204 (15)
3.5
d
6
1979/02/10
694 (75)
20
b,c,e
7
1981/12/30
174 (5)
13
a,d
8
1983/11/18
164 (1)
200
a,d
9
1987/02/25
52 (1)
2
a,d
10
1989/11/22
164 (15)
2
a,d
11
Predictive value of
susceptibility classes
Susceptibility
class
I
II
III
IV
V
Critical thresholds of rainfall
(quantity-duration)
responsible for landslide events
Landslide
type B
217 (15)
4.5
a,d
12
1989/12/05
333 (30)
5.5
b,c,e
13
1989/12/21
1989/11/25
495 (40)
20
b,c,e
14
1996/01/09
544 (60)
10
b,c,e
%
15
1996/01/23
686 (75)
18
b,c,e
41
29
16
14
0
16
1996/01/28
495 (40)
20
b,c,e
17
1996/02/01
793 (90)
24
b,c,e
18
2001/01/06
447 (60)
5
c
19
2001/01/09
467 (60)
5.5
c
1956 - 2001
Landslide typology:
a) shallow translational slides
b) deep translational slides
c) rotational slides
d) slope movements triggered
by bank erosion
e) complex and composite
slope movements
9
Cumulative rainfall (mm)
Three scenarios for future landslide activity within the
Fanhões-Trancão test site based on past landslide events
Rotational slides
Translational slides
Complex slope movements
Shallow translational slides
Slope movements
triggered by bank erosion
900
800
700
600
500
400
300
200
100
0
Critical rainfall
amount/duration
(mm/days)
Scenario
0
20
40
60
80
Return
Period
(years)
Affected area by
shallow translational
slides (m2)
(1) 1979, February
128/3
8.5
44,440
(2) 1983, November
164/1
200
47,125
(3) 1989, November
217/15
4.5
1,315
100
Duration
(consecutive
days)
Duration
(days)
landslides
no landslides
Cr = 6.3D + 70
Age of and total affected areas by shallow translational slides
within the Fanhões-Trancão test site.
Age
N
(%)
Total affected area (m2)
(%)
1967 or prior
22
22.0
44,017
31.0
1979, February
24
24.0
44,440
31.3
1983, November
40
40.0
47,125
33.1
1987, February
4
4.0
3,283
2.3
1989, November
3
3.0
1,315
0.9
1989, December
3
3.0
996
0.7
1996, January
4
4.0
1,000
0.7
HAZARD ASSESSMENT AT A
PROBABILISTIC BASIS
For each particular triggering scenario, the conditional
probability that a pixel will be affected by a shallow
translational slide in the future is estimated by:
⎛ Taffected
⎞
P =1 − ⎜⎜1 −
. pred ⎟⎟
Ty
⎝
⎠
Where:
Taffected = Total area to be affected by landslides in a scenario (x);
Ty = Total area of susceptibility class y
Total
100
100.0
142,176
100.0
pred = prediction value of susceptibility class y.
Calculation of probabilities for landslide hazard assessment
working on a scenario basis
General assumption:
Probability to each pixel to be
affected by a landslide
Scenarios
The rainfall patterns (quantity/duration) which
produced slope instability in the past will
produce the same effects in the future (i.e.
same types of landslides and same total
affected area).
Landslide
susceptibility
class
Area (number
of pixels)
(pixel= 5m)
Predictive value of
susceptibility class
(1)
February
1979
(2)
November
1983
(3)
November
1989
I - Top 8%
64150
0.4071
0.0113
0.0120
0.00034
II – 8-18%
82044
0.2885
0.0062
0.0066
0.00019
III - 18-35%
142342
0.1647
0.0021
0.0022
0.00006
IV - 35-84%
382114
127459
0.1397
0.0006
0.0000
0.0007
0.00002
0.00000
V – 84-100%
0.0000
0.0000
10
Shallow translational slides
Fanhões-Trancão test site
ROTATIONAL
SLIDES
Joint Conditional Probability
Function
Prediction-Rate Curve
Hazard map
Triggering scenario (1)
Percentage of predicted
validation group landslides
100
I
II
III
IV
V
III
80
70
V
II
60
50
40
30
20
I
10
0
128 mm / 3 days
(RP=8.5 years)
Class
IV
90
0
10
20
30
40
50
60
70
80
90
100
Percentage of study area predicted as susceptible using
estimation group landslides
Predictive value of
susceptibility classes
Probability
by pixel
0.0113
0.0062
0.0021
0.0006
0.0000
Shallow translational slides
Fanhões-Trancão test site
Susceptibility
class
I
II
III
IV
V
%
13
44
11
32
0
TRANSLATIONAL
SLIDES
Joint Conditional Probability
Function
Hazard map
Triggering scenario (3)
217mm / 15 days
(RP=4.5 years)
Percentage of predicted
validation group landslides
Prediction-Rate Curve
100
90
80
70
60
50
40
30
20
10
0
II
III IV
V
I
0
10
20
30
40
50
60
70
80
90
100
Percentage of study area predicted as susceptible using
estimation group landslides
Class
I
II
III
IV
V
Probability
by pixel
0.00034
0.00019
0.00006
0.00002
0.00000
Predictive value of
susceptibility classes
Susceptibility
class
I
II
III
IV
V
%
61
14
4
21
0
Shallow translational slides
Fanhões-Trancão test site
Joint Conditional Probability
Function
House
(500 m2)
Hazard map
⎛
⎞
Taffected
P ' = 1 − ⎜⎜1 −
.0.41(= predI )⎟⎟
64,150(= TI )
⎝
⎠
20 (= size of house )
Taffected = Total area to be affected by landslides
in a scenario (x);
TI = Total area of susceptibility class I;
PredI = prediction value of susceptibility class I.
Probability that a part of
the house will be involved
in landslide activity
Scenario (1)
Scenario (2)
Scenario (3)
WORKING ON A SCENARIO BASIS
Scenario January 1996:
Critical rainfall amount/duration - 495mm/40 days
Return period - 20 years
Affected area by landslides (m2)
Rotational slides
48,127
Translational slides
6, 552
Shallow translat ional slides
1, 000
20.4%
21.5%
0.7%
11
Landslide
susceptibility
class
ROTATIONAL
SLIDES
I - Top 1%
II - 1-12%
III - 12-17%
IV - 17-70%
V - 70-100%
Landslide
susceptibility
class
TRANSLATIONAL
SLIDES
SHALLOW
TRANSLATIONAL
SLIDES
8122
88934
40334
423765
236954
Probability to
Predictive value
each pixel to be
of susceptibility
affected by a
class
landslide
0.6080
0.0022
0.1418
0.0011
0.0426
0.0003
0.2076
0.0001
0.000
0.0000
Area (#
pixels)
I - Top 9%
II - 9-13%
III - 13-17%
IV - 17-75%
V - 75-100%
72518
32423
33071
465020
195077
Landslide
susceptibility
class
Area (#
pixels)
I - Top 8%
II - 8-18%
III - 18-35%
IV - 35-84%
V - 84-100%
Probability to
Predictive value
each pixel to be
of susceptibility
affected by a
class
landslide
0.1275
0.0302
0.4448
0.0096
0.1129
0.0054
0.3148
0.0014
0.000
0.0000
Area (#
pixels)
Probability to
Predictive value
each pixel to be
of susceptibility
affected by a
class
landslide
0.4071
0.00025
0.2885
0.00014
0.1647
0.00005
0.1397
0.00001
0.000
0.00000
64150
82044
142342
382114
127459
TOWARDS LANDSLIDE RISK ASSESSMENT
AND MANAGEMENT
Temporal
dimension
Susceptibility
Causes
Hazard
Vulnerable elements
Effects
Typology
Intensity
Value
Vulnerability
Potential loss
Specific Risk
TOTAL RISK
RISK MANAGEMENT
Acceptable Risk
adapted from Canuti & Casagli (1994 )
12