02 mountainrisks intensivecourse barcelona 08 vanwesten susceptibility
DESCRIPTION
Free Disaster mannagementITCTRANSCRIPT
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Landslide susceptibility assessmentLandslide susceptibility assessment
Cees van WestenUnited Nations University ITC School for Disaster Geo-Information Management
International Institute for Geo-Information Science and Earth Observation (ITC)Enschede, The Netherlands
E-mail: [email protected]
Associated Institute of the
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
SuceptibilitySuceptibility: contents of lecture: contents of lecture What is susceptibility What controls it:
Landslide type Scale Objectives
Which data are required? Which techniques can be used to assess it?
Heuristic, statistical, deterministic Validation From susceptibility to hazards Discussion part..Which method to use when?
Landslides Susceptibility
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
What is susceptibility?What is susceptibility?Landslide susceptibility is: The relative spatial likelihood for the occurrence of
landslides of a particular type and volumeLandslide hazard is: The probability of occurrence of a particular landslide
type (initiation and run-out, volume, speed) within a specified period of time and in a given area.
Landslide risk is: The expected losses (monetary, or in number of buildings
/ people) due to specific landslide type type (initiation and run-out, volume, speed) within a specified period of time and in a given area.
Where?
Spatial probability
When?
Temporal probability
Hazard = Susceptibility * Triggering factors
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Population, building, industry, agriculture,Infrastructure, services
Landslides Inventory
Spatial landslide risk assessment frameworkSpatial landslide risk assessment framework
OccurrenceEnvironmentalParameters Triggering Factors Elements at Risk
Susceptibility
Natural HazardSpecific Risk
Total Risk
Vulnerability
Geology, Soil,Landuse,Slope, HeightInternal relief
Earthquakesand
Rainfall
Susceptible areas for the initiation of landslide.
Probability of occurrence within a specified period of time and within a given area of a landslide.
Expected degree of loss due to landslide.
Degrees of loss to a given element(s) at risk resulting from the occurrence of a landslide of a given magnitude. [0, no loss.. 1, total loss]
Expected number of lives lost, persons injured, damage to property, or disruption of economic activity due to landslide.
type, magnitude,time, activity
values, classes time, magnitude,intensity
object and attributes
FrequencyAnalysis
(type, magnitude)/ time
Run-outValue
%- (probability) /(type, magnitude)
% / (type, magnitude) / time
FrequencyAnalysis
(intensity, magnitude)/ time
$ / object[0..1] / (type,magnitude, distance)
% / [0..1] / (type, magnitude) / time
% / $ / (type, magnitude) / time
Number of occurrence in a given time.
Number of occurrence in a given time.
Landslide hazard methods
Heuristic analysis
Statistical analysis
Inventory analysis
Deterministic analysis
Consequence
$ / (type, magnitude)
Loss outcome from the occurrence a landslide of certain type or magnitude.
Risk evaluation
Acceptable risk
Tolerable risk
m - (distance) /(type, magnitude)
Costs of building, engineering works, infrastructure, environmental features, and economic activities in the area affected by landslides.
Potential path covered by the landslide occurrence.
E. C
astellanos,, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
What is required for expressingWhat is required for expressing
RealRealYesYes
AbsoluteAbsoluteYesYes
RelativeAt best relativeYesRelative
Landslide initiationSpatial probabilityTemporal probabilityLandslide typeVolume
YesYesYesYesYes
YesYesYesYesYes
NoNoNoNoNo
Landslide runoutSpatial probabilityTemporal probabilityLandslide typeVolumeSpeed
YesYesYes
NoNoNo
NoNoNo
Landslide consequencesBuilding damageInfrastrusture damagePopulation damage
Component RiskHazardSusceptibility
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
What controls landslide susceptibility?What controls landslide susceptibility? Landslide distribution
Landslide types Failure mechanisms Age & activity Area covered by landslides Controlling factors Type
Homogeneity Triggering factors
Earthquakes Rainfall Human interventions Others
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
All landslides knownAll landslides known??
Red : landslide initiation on cut slope Pink : Landslide run out Dark green : landslide initiation on natural
slope Light green : Landslide run out Triangles : Rain gauge location
Type of study area? Watershed Line Point
Basically all landslides are known from the past 30-40 years based on registers
Pankaj Jaiswal, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
An example of rapid change in land cover in the Nilgiri hills. Marapallam debris slide as seen in different years.
1993 1998 2007
All landslides known?All landslides known?
Pankaj Jaiswal, 2008
Mapping landslides from images without prior knowledge is a tricky business many old landslides are not properly known
The use of (as many as possible) sets of older images (airphotos and satellite images improves the interpretability.
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Who needs a susceptibility map if you know all slides
CEO Hong Kong: Enhanced Natural Terrain Landslide Inventory (ENTLI) : ~15,800 recentlandslide features have been identified based on airphoto interpretation. For each terrain unit the temporal and spatial probability is known.
Ken Ho, CEO, HongKong, 2007
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Scale of analysisScale of analysis National scale
(< 1:1.000.000) Mainly inventory. Public awareness & policy support
Regional scale(1:100.000 - 1:1.000.000) For reconnaissance phases for planning projects for the construction of infrastructural works, or agricultural development projects.
Medium scale(1:25.000 - 1:100.000). For land use planning and construction of infrastructural works, environmental impact assessment and municipal planning.
Large scale(1:2.000 - 1:25.000). For risk assessment and detailed planning.
Site investigation(>1: 2.000). Detailed risk assessment, and design of slope stabilization works
Geological Periode.g. Middle Eocene: Aleutorite, marl, limestone, chert, conglomerate, olitostrome.
Geological formationse.g. Yateras formation: limestones, marls and clays
Lithological unite.g. karstic limestones
Geotechnical parameterse.g. Internal friction 5, bulk unit weight 16.5 kPa/m, viscosity 2.5 kPa.s
GEOLOGY
National level1:1,000,000 scale
Provincial Level1:100,000 scale
Municipal level1:50,000 scale
Local level1:25,000 scale
Np
Nmh
Nb
Ha
Hazard example
Castellanos and Van Westen, 2008
Each scale has its own objectives, and has its own possibilities for data
collection
-
International Institute for Geo-Information Science and Earth Observation (ITC)
1 : 2,750,000
Landslide Susceptibility in GermanyLandslide Susceptibility in Germany
93,8
5,4
0,6
0,2
Very lowLowModerateHigh
Building destruction likely, people endangeredHigh
Building damage possible, people probably endangeredModerate
Building damage and directly affected people unlikelyLow
At human discretion no dangerVery low
DescriptionClass
Glade, Dikau and Bell
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Example: CubaExample: Cuba
Local disaster risk reduction plan Identifying elements at risk affected by different scenarios and implementing mitigation actions
Local and municipal government and civil defence
QuantitativeRunout modelling based on geotechnical parameters(adaptation needed for generalization)
Local(1:25,000)
Municipal disaster risk reduction planEstimating losses and delimitating areas for mitigation actions
Municipal government and civil defence
QuantitativeTMU and expert judgment(adaptation needed for generalization)
Municipal(1:50,000)
Provincial disaster risk reduction planLocating landslides and priority areas for investigating causes and consequences
Provincial government and civil defence
Semi-quantitative with SMCESpatial model viaStatistical method and weighting
Provincial(1:100,000)
Locating priority areas for regional studies and guide national policyPeriodic assessment to monitor local improvement
National Civil Defence
Semi-quantitative with SMCERisk index by ranking and weighting
National(1:1,000,000)
UseOrganizationMethodLevel (scale)
Castellanos and Van Westen, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Geotechnical parametersHydrological parameters
Geotechnical parametersHydrological parameters
Hydrological parameters
Slope hydrology
R
Scale
M LUse
RS
H
DEM Slope / Aspect etcElevation
Building footprintsPopulation data
Infrastructure
Land use
Catalogs & strong motion
Hazard maps
Rainfall / Temperature
Flow accumulation
Detailed mapping units
Main units
Soil depth
Soil types
Faults
Structure
Lithology
Occurrence/Inventory
Elements at risk
Seismic data
Meteo data
Geomorphology
Hydrology
Soils
Geology
Landslides
ModelsUpdate Frequency
>10 DPSday0.8110
High
Moderate
Low
Input dataInput dataRegional Medium Large
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Which data to collect?Which data to collect?Objectives of the study
Availability of existing dataAvailable resources
Complexity of the area
Selection of susceptibility analysis technique
Selection Collection spatial and non spatial data
Type of landslidesSize of study area
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Main methodsMain methodsKnowledge driven: Boolean Logic Fuzzy logic Multiclass overlay Spatial multicriteria
evaluationData driven: Bivariate statistics Weights of evidence Information value Frequency ratio Multi-variate statistics Logistic regression Discriminant analysis Cluster analysis Artificial Neural Networks
Deterministic methods Static methods Infinite slope based Profile based Dynamic methods
Inventory based Magnitude-frequency Activity mapping
Probabilistic methods Parameter uncertainty Temporal prediction
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Landslide activity analysis
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Landslide mapping can be difficultLandslide mapping can be difficult
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Direct / Indirect methodsDirect / Indirect methods Direct methods
experience driven applied geomorphological approach, where the earth scientist evaluates the direct relationship between the landslides and the geomorphological and geological setting during the survey at the site of the failure.
Indirect methodsthe mapping of a large amount of parameters and the (statistical or deterministic) analysis of all these possible contributing factors in relation to the occurrence of slope instability phenomena, determining in this way the relation between the terrain conditions and the occurrence of landslides. Based on the results of this analysis statements are made regarding the conditions under which slope failures occur.
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Geomorphological hazard mappingGeomorphological hazard mapping
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Storing Storing GeomorphologicalGeomorphological data in GISdata in GIS Store Geomorpholo-
gical information in several data layers:A: Main UnitsB: Sub unitsC: LandslidesD: Material typesE: Hazard
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Geomorphological hazard mapGeomorphological hazard map
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Rating methodsRating methods Index methods: predefined,
e.g. Mora-Vahrson method
(Central America) BIS method (India) Problem: they are either very
general and cannot be applied in larger scales
Differences between areas are ignored
Sometimes do not even require a landslide inventory
Index methods: experts derived
Index overlay, multi-class overlay
Fuzzy-logic Spatial Multi Critera
Evaluation
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Knowledge driven methodKnowledge driven methodThe expert decides: On which criteria are used On which input maps are selected On the weights of the classes of
the input maps On the weights of the input maps
themselves
Subjective Not reproducible Difficult: how important are the
different factors
Is there a way to improve and make this more objective?
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Spatial MultiSpatial Multi--Criteria EvaluationCriteria Evaluation
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
The criteria treeThe criteria tree
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
The criteria treeThe criteria tree
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Multi Criteria EvaluationMulti Criteria EvaluationInventory Environmental Parameters
Geomorphometric Ground ConditionsTriggering Factors
National LandslideMitigation PlanNational LandslideMitigation Plan
Data sources: SRTM DEM derivatives maps (90m) National Atlas of Cuba (1:1,000,000 scale) National Statistics Office (ONE) National Organizations
Area: 110,860 km2Analysis unit: pixel based at 90 m.Risk assessment method: semi-quantitative spatial multicriteria.Final map: at 1:1,000,000 scale
Internal relief
Landslides
Slope Angle
Geology
Landuse
Earthquakes
Rainfall
Provinces
Municipalities
SusceptibilityIndex
a. slope angle, b. geology, c. landuse, d. maximum rainfall
expected in 24h and e. maximum peak ground acceleration
Castellanos and Van Westen, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Final output mapFinal output map
Castellanos and Van Westen, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Knowledge driven methodsKnowledge driven methodsAdvantages: Expert knowledge can be properly included Flexible, both in causal factors, as in sub-goals Can be used for both hazard and risk assessmentDisadvantages: Subjectivity (however.) Doesnt give quantitative estimation of hazard Cannot be used for quantitative risk assessment
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Data driven methodsData driven methodsThe expert decides: On which criteria are used On which input maps are selected
Weights are determined: By comparing the inputs maps
with known occurrence of the feature one would like to predict
Basically based on densities Converted into spatial
probabilities Using bivariate or multivariate
methods Results can be validated
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
BivariateBivariate statistical methodsstatistical methods Frequency ratio:
Area of landslides in Class / Area of all LandslidesFR =
Area of Class / Entire map
Area of landslides in Class / Area of ClassFR = LN
Area of all landslides in the map / Area of Entire map
Hazard index:
Weights of evidence:i+
ei
iW =
P { B |S}P { B |S }
log
i-
ei
iW =
P { B |S}P { B |S }
log
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Slope angle versus Slope angle versus landslideslandslides
SRTM DEM
1:50000 topomap
1:2000 topomap
Lidar DEM
Influence of input dataInfluence of input data
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
BivariateBivariate statisticalstatistical analysisanalysis
(Castellanos & van Westen, 2008)
Soil
Cross tablesFactor maps Weight maps
M
a
p
o
v
e
r
l
a
y
i
n
g
p
r
o
c
e
d
u
r
e
Slopes
Landslides
Geology
Susceptibility map
WSoils
WSlopes
WGeology
C
o
m
b
i
n
a
t
i
o
n
o
f
w
e
i
g
h
t
s
.
.
.
.
.
....
E. Castellanos, Jun-2008
Validation
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
ExampleExampleB B
S
S
npix1 npix2
npix3 npix4
180 20 200
3420
3600 6400 10000
6380 9800
i+
ei
iW =
P { B |S}P { B |S }
log
i-
ei
iW =
P { B |S}P { B |S }
log
i+
e
1
1 2
3
3 4
W = +
+ Npix
log
NpixNpix Npix
NpixNpix
i-
e
2
1 2
4
3 4
W = +
+ Npix
log
NpixNpix Npix
NpixNpix
=
180180 + 20
3420 + 63803420
=0.9
0.349= Ln 2.578 = 0.9474
=
20180 + 20
3420 + 63806380
=0.1
0.651= Ln 0.1536 = -1.87
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)Example of statistical analysis Evidence maps
drain fault roadgeomgeolAspcl slidelanduseslope slide60
Rasterize; distance calculation
Create class, group domain, Reclassify
Class_distroadClass_disrtdrainClass_distfault
Create attribute map, calculate prior probability for all landslides in study area. Then, create a bit map showing only
active landslides
bslide60bslide
Cross operation between factor maps and evidence maps
Waspclw60aspcl
Wgeomw60geom
Wlandusew60landuse
Wslopew60slope
Wfaultw60fault
Wdrainw60drain
Wroadw60road
Wgeolw60geol
To calculate four possible combinations of potential landslideconditioning factor and a landslide inventory map that are npix shows
pixel number. Then calculate weight of evidence can be written in number of pixels as follow equations:
i+
e
1
1 2
3
3 4
W = +
+ Npix
log
NpixNpix Npix
NpixNpix
i-
e
2
1 2
4
3 4
W = +
+Npix
log
NpixNpix Npix
NpixNpix
To create final weight evidence maps for each evidence maps (Use script)To calculate success rate and predict rate then reclassify final weight evidence map.
To compare the result map, which is shows added activity landslide. Also this map can be to check result is how well predicting final weight evidence map.wfinal w60final
Prediction rate Success rate
Prediction map of landslides
Class_wfinal Class_w60finalAdded landslide area
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Combine methods..Combine methods..
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Landslide mappingLandslide mapping
100.0281Total
6.418Topples
76.5215Slides
7.822Rock fall
9.326Debris flow%Frequency
Castellanos, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Hazard analysis Hazard analysis -- factorsfactorsGeomorphology Geology Landuse
Slope angle Slope aspect Internal relief
Drainage density Soils
Rainfall PGA Road distance
Fault distance
Castellanos, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
MultiMulti--variatevariate modelsmodels Multiple regression Logistic regression Artificial Neural Network Basic units:
Pixels Unique condition polygons DTM slope units
Castellanos, Melchiore and Van Westen, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Susceptibility maps per causal mechanism Susceptibility maps per causal mechanism Debris hazard
No hazardLow hazardModerate hazardHigh hazard
SlideT hazard
Rockfall hazard Slide hazard
Topples hazard All hazards
Castellanos and Van Westen, 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Combine methods..Combine methods..
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Main data types: Soil data
Depth Hydrological properties Geotechnical properties
Digital Elevation Models Slope angle Local drain direction Contributing areas
Groundwater measurements
Physically based modelingPhysically based modeling
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Deterministic Models: No Uncertainty (Single value at a given space& moment of time)
Stochastic Models: Involves Uncertainty (Probability depended)
Static Models: Independent of time. Time Frozen Dynamic Models: Simulates changes through time
Process Driven Models: Models based on general physical laws. Data Driven Models: Models valid under certain conditions defined
by the properties and processes of the study area.Eg: Downward approach to hydrological model development
Physically based modelingPhysically based modeling
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Dynamic modelsDynamic models
IO: Institute of Origin; PE: Programme Environment; TC: Temporal Component; D: Dimension; HC: Hydrological Component; SPC: Spatial Parameterization Capability
IncapableCapable (not recommended)Capable (not
recommended)Fully capableSPC
Wetness Index [Bevenand Kirkby, 1979]
Wetness Index [Beven and Kirkby,
1979]
Infiltration, Runoff,
Pressure head at min FOS
Water Level, Effective Degree of Saturation,
Runoff, Bed Rock StorageHC
1DStatic
ArcView
University of California (Berkeley),
and Stillwater Sciences
SHALSTAB
1DStatic
ArcView
Utah State University
SINMAP
2 D + 1DDynamicFortran
USGS
TRIGRS
2.5 D + 1DDynamicPCRaster
Utrecht University
STARWARS + PROBSTAB
D
TC
PE
IO
Different implementations of the infinite slope model
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
STARWARS + PROBSTABSTARWARS + PROBSTAB
STARWARS
DischargeBRStore
WL
DEMSD
CRC BU S
PROBSTAB
ETRF Ksat hA n
PFFOS
RF: Rainfall (time series) (m/d) ET: Evapotranspiration (m/d) Ksat: Saturated Hydraulic
Conductivity (m/d) hA: Air Entry Value (m) : Slope of SWRC (-) N: Porosity (-) RC: Root Cohesion (kPa) C: Soil Cohesion (kPa) : Internal friction angle BD: Bulk Unit Weight (kN/m3) S: Surcharge (kPa) BRStore: Bed rock store (m3) WL: Water level (m) : Volumetric moisture content PF: Probability of Failure FOS: Factor of Safety
Perched Water Level (m)Perched Water Level (m) Factor of Safety (with Factor of Safety (with root cohesion)root cohesion)
Factor of Safety (withFactor of Safety (without root cohesion)out root cohesion)
Sekhar et al., 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
NRI: Net Rainfall Intensity (time series) (m/sec)
Ksat: Saturated Hydraulic Conductivity (m/sec)
C: Cohesion (kPa) : Internal friction angle () uwv: Unit Weight of Water (Kgm/s2) uws: Unit Weight of Soil (Kgm/s2) idw: Initial depth of water (m) inf: Initial infiltration rate (m/sec) D: Diffusivity (m2/sec) SD: Soil depth (m) PP: Pressure head at maximum depth FOS: Factor of Safety (-)
uww uws SD DEMinfidw D Ksat C
NRI
TRIGRS
FOSPP
TRIGRSTRIGRS
FOS
Factor of Safety (with Factor of Safety (with root cohesion)root cohesion)
Factor of Safety (withFactor of Safety (without root cohesion)out root cohesion)
Sekhar et al., 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
StabilityIndex
Phimax Phimin T/Rmax T/Rmin Cmax Cmin
DEM SINMAP
Scenario actual inputs
Scenario default inputs
SINMAPSINMAP
Scenario default inputs
Scenario actual inputs
Sekhar et al., 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
StabilityIndex
DEM SHALSTAB
Phi q/t CSden
Scenario default inputs
Scenario actual inputs
Scenario default inputs Scenario actual inputs
SHALSTABSHALSTAB
Sekhar et al., 2008
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Model comparisonModel comparison STARWARS+PROBSTAB provided better results than other modelsSTARWARS+PROBSTAB provided better results than other models
99 Area of failure over estimatedArea of failure over estimated99 Temporal failure conditions predicted fairly wellTemporal failure conditions predicted fairly well
SHALSTAB provided better results than SINMAPSHALSTAB provided better results than SINMAP99 Not to over look the capability of SINMAP for distributed Not to over look the capability of SINMAP for distributed
parameterizationparameterization Applicability of TRIGRS is limited in data poor regions because Applicability of TRIGRS is limited in data poor regions because of:of:
99 Lack of rainfall intensity data in most landslide hazard prone aLack of rainfall intensity data in most landslide hazard prone areas of reas of the developing worldthe developing world
99 Lack of knowledge of initial conditionsLack of knowledge of initial conditions
8
2.8
SHALSTAB(limited to stability index
-2.5)
13
3.7
SINMAP(limited to lower
threshold)
1518Known landslide locations
predicted failed during the critical season (FOS
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
ValidationValidation
Best validation is to wait and see Very few publications on checking landslide
hazard maps (or risk maps) some time later. Now it is based on historical data. Dynamic deterministic modeling allows to
validate when and where Other methods only where.
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
ValidationValidation Several studies have been done on comparing
method Very few studies have been done on checking old
landslide maps.
Partition of the data setEstimation Group and Validation Group
Methods of Crossvalidation
- random validation
- spatial validation (moving window)
- temporal validation
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Success & Prediction rateSuccess & Prediction rateP
e
r
c
e
n
t
a
g
e
o
f
a
l
l
l
a
n
d
s
l
i
d
e
s
Percentage of weight map ordered from high to low
70 percent of all new landslides is located in 30 % of the map with highest prediction scores
Success rate: How well the model performs.Prediction rate: How well the model predicts
(Chung & Fabbri, 1999)
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Success rateSuccess rate
ANN with continuous variables
ANN with classes variables
Weight of evidence all variables
Weight of evidence combining variables
Topples
Slides
Large rockslides
Rockfalls
Debrisflows
Castellanos, Melchiore and Van Westen, 2008
If you use heuristic analysis, the success rate could be very high
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Results..Results..
91.7 %1 %
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
From susceptibility to hazardFrom susceptibility to hazard
4426 buildings
9645buildings
22019buildings
Risk = Hazard * Vulnerability * Amount
How much percentage of the high, moderate and low hazard classes may be affected by landsides?
In which period will these landslides occur? What is the vulnerability to landslides?
Results using mapping unitsHigh Moderate Low
Known nowStill to doOnly susceptibility
Hazard = Spatial probability * Temporal probability
The temporal probability that landslides may occur due to a triggering event. Here we will link the return period of the triggering event with the landslides that are caused by it. We have differentiated return periods of: 50, 100, 200, 300 and 400 years.
The spatial probability that a particular area would be affected by landslides of the given temporal probability. This is calculated as the landslide density within the landslide susceptibility class.
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
From susceptibility to hazardFrom susceptibility to hazardMillion dollar information!!!
Landslide related to different return periods
Landslide_ID map If the indication of the high, moderate and low areas susceptibility is correct, different landslide events with different return periods will give different distributions of landslides in these classes.
The probability can be estimated by multiplying the temporal probability (1/return/period for annual probability) with the spatial probability (= what is the chance that 1 pixel is affected)
Susceptibility
Cross
Density in high
Density inmoderate
Density inlow
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Calculating hazardCalculating hazardAssumption is that events with a larger return period
will also trigger those landslides that would be triggered by events from smaller return periods
S
u
s
c
e
p
t
i
b
i
l
i
t
y
c
l
a
s
s
e
s
Return periods
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
From susceptibility to hazardsFrom susceptibility to hazards
-
International Institute for Geo-Information Science and Earth Observation (ITC)
ConclusionsConclusions
No
No
Yes
Process-basedAnalysis
Quantitative methodsQualitative methods
NoYes/NoYesYes> 1:100,000
YesYesYesYes1:25,000 1:50,000
YesNoYesYes< 1:10,000
Neural networkAnalysis
StatisticalAnalysis
HeuristicAnalysisInventoryScale
based on Soeters & van Westen (1996) and Aleotti & Chowdhury (1999)
Relation between scale and landslide susceptibility models
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
ConclusionsConclusions Separate landslide susceptibility models
are needed: For different landslide types For different failure mechanisms
Sometimes it is better to skip susceptibility assessment. If you already know the landslide initiation areas If landslides are caused by reactivation of old ones
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Conclusions: methods..Conclusions: methods..
There are many papers that compare methods for susceptibility assessment
Susceptibility research is much more model driven, than data driven.
Much more emphasis should be given on the analysis of appropriate input data, and of the inclusion of expert opinion in the analysis.
Most methods report that around 75% of the landslides were predicted. What about the 25%?
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Conclusions: dataConclusions: data The availability of data greatly determines the
possibility of using particular susceptibility methods
Physical based models require detailed soil/rock characteristics
Depends on the homogeneity of the area Dont simple use the factors everyone is using:
GIS-based factor derived from DEMs are easy to make but are they also useful?
Strong relation between failure mechanism and factors to collect.
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
ConclusionsConclusions Validation.
A susceptibility map is useless unless it is validated. Spatial validation Temporal validation
Validation of statistically derived results are often more difficult
How long is a susceptibility map valid? As soon as any of the (intrinsic or triggering) factors
changes, e.g.: Road construction Climate change
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
RiskCityRiskCityTraining package on landslide Training package on landslide
susceptibility, hazard and risk assessmentsusceptibility, hazard and risk assessment
-
ISL 2004
International Institute for Geo-Information Science and Earth Observation (ITC)
Discussion:Discussion:
What would be the best methods to use in the Mountain risk study areas? Valtellina? Firenze? Dolomites? Andorra? Barcelonette? Swiss areas?
= landslide = susceptibility