GIS in Geology - milosmarjanovic
Download
Report
Transcript GIS in Geology - milosmarjanovic
GIS IN GEOLOGY
Lesson 5
4.11.2010.
Miloš Marjanović
GIS in Landslide assessment (advanced)
Statistical analysis of landslide susceptibility/hazard/risk zonation
1.
Comparing landslide occurrence from inventory or on-the-site data and input
parameter relevance (weight, or rank according to the density of parameter
classes) in the final model by different techniques of statistical dependancy
Deterministic models for landslide susceptibility/hazard/risk zonation
2.
Coupling slope stability criteria (static equilibrium) and triggering factor(s)
influence(s) in order to map where (& when) the triggering factor of certain
intensity overcomes the soil/rock strength, causing the slope failure
Accent on advances in modeling approaches as research level
upgrades and upscales
GIS in Landslide assessment (advanced)
Computer Aided
Drawing (CAD)
Database
Management
Systems
(DBMS)
Image
Processing (IP)
Geographic
Information System
(GIS)
Artificial
Intelligence(AI)
Desktop
mapping
Desktop and
Web publishing
General
statistics
Spreadsheets
Contouring and
surface modeling
Geostatistics
GIS in Landslide assessment (advanced)
Once gain the procedure of susceptibility/hazard/risk zoning
Preparation, adjusting scale and level of research
Input parameters
Performing susceptibility zonation by combining the inputs in knowledge
(as presented in Lesson 3) or data driven approaches over training sets
Calibration over testing sets
Selecting the best models with the smallest errors
Shifting from susceptibility to hazard and risk
Additional inputs for frequency analysis (spatial-temporal probabilities)
Implementing element at risk by thematic maps (population, infrastructure, dwelling) of ER
vulnerability
Appending upon previous susceptibility map trough risk equation, R=H*V(ER)
GIS in Landslide assessment (advanced)
Statistical techniques of landslide susceptibility/hazard/risk zonation
(applicable from regional to slope scale)
1.
Bivariate
Multivariate
Discriminant score
Logistic regression
Cluster Analysis
Principal Component Analysis (PCA)
Machine learning (advanced statistical approach)
Artificial Neural Networks
Support Vector Machines
Decision Trees
Fuzzy Logics
GIS in Landslide assessment (advanced)
Bivariate statistics
Relating two maps using descriptive statistics
Procedure:
1.
Overlaying i-th geo-parameter map and landslide
reference map, calculating landslide density per each
class and overall landslide density
2.
Calculating the weight per each class by relating
class to overall density
3.
Reclassification of initial geo-parameter map
4.
Combination of geo-parameter maps into a final map
5.
Reclassify the final map into levels adjusted by initial
landslide map
Techniques:
Information value
Weights of evidence
Frequency ratio
GIS in Landslide assessment (advanced)
Bivariate statistics techniques
Information value
Weight relates densities of landslide per class and per entire map
Wi ln
class
# Lpxclass # pxclass
ln
overall
# Lpxclass # pxclass
Calculate +/– weights (how important is the presence/absence of geo-parameter class in the
landslide reference map)
W+=0 no contribution effect (irrelevant factor)
W– =0 no contribution effect (irrelevant factor)
W+>0 contributes the presence of landslides
W–>0 contributes the absence of landslides
W+<0 contributes the absence of landslides
W–<0 contributes the presence of landslides
Repeat per every geo-parameter (geology, slope, land cover, elevation…)
Calculate probability of landslide occurrence:
PROB Alandslides Aclass
GIS in Landslide assessment (advanced)
Bivariate statistics techniques
Weight of evidence
Weight relates densities of landslide per class and per entire map
Wi ln
class
# Lpxclass # pxclass
ln
overall
# Lpxclass # pxclass
Sum-up +/– weights
W=0 no contribution effect (irrelevant factor)
W>0 contributes the presence of landslides
W<0 contributes the absence of landslides
Repeat per every geo-parameter (geology, slope, land cover, elevation…)
Calculate probability of landslide occurrence:
PROB Alandslides Aclass
GIS in Landslide assessment (advanced)
Multivariate statistics
Relating all geo-parameters (independent variables) to reference landslide map
(dependent variable) simultaneously with correlation between the independent
variables
Procedure:
1.
Quantification and normalization of the inputs (note that with bivariate categorical classes
were possible)
2.
Group independent variables in classes as in bivariate case
3.
Correlate the input variables between each other by bivariate correlations or AHP or black
box models (AI approach)
4.
Solve the distribution in a hyper-plane that separates the initial cluster of data
Techniques:
Discriminant score
Logistic regression
Cluster analysis
GIS in Landslide assessment (advanced)
Multivariate techniques
Discriminant score
If certainathreshold
Assumes
distribution
is reached
betweenthe
theDS
parameters
function is to be
classified andand
appropriate
divides
it could
them
serve
in two
theclasses:
model stable A and
unstable
AcceptedBweight factors are used to generate the final
Generate
model
of susceptibility/hazard/risk
a geo-parameters relation table
Compare results
Interrelates
all the
according
inputs bytoDiscriinant
the susceptibility
Score function:
index
with
other
DS=A
+A methods
P +A P +…+A P
0
1 1
2 2
n n
where Ai is the overall weight factor in the score
Pi is the parameter (geology, slope, elevation…)
Project a hyper-plane to discern classes A and B
GIS in Landslide assessment (advanced)
Multivariate techniques
Machine learning algorithms
K-Nearest Neighbor (KNN)
Votes per unclassified point
Hardware demanding (sorting + voting) and therefore trained on small sets
Convenient for spatially correlated data
(clustered data)
Support Vector Machines (SVM)
Separates classes by plane with the widest margin
If that plane could not be set in ordinary dimension space (2-3D)
it is plotted in higher feature space where observed set is projected
by kernel function (Gaussian)
Training set could be significantly reduced with high quality of data
GIS in Landslide assessment (advanced)
2.
Deterministic models for landslide susceptibility/hazard/risk zonation
(applicable from regional to local scale):
SHALSTAB: parametric free, simple hydrologic model, shallow landsliding, steady
state
TOPOG: additional soil parameters, simple hydrologic model, shallow landsliding,
steady state
SINMAP: additional soil parameters (uncertainty included), simple hydrologic
model, shallow landsliding, steady state
TRIGRS: advanced 1-D hydrologic model, shallow landsliding, steady state
GeoTOP: advanced 3-D hydrologic model, shallow landsliding, steady state
DYLAM: requires geo-mechanical and meteorological inputs, simple hydrologic
model, shallow landsliding, dynamic
GIS in Landslide assessment (advanced)
SHALSTAB (SHAllow Landslide STABility)
Concept: couple the slope stability and hydrologic model
Triggering mechanism: atmospheric discharge (heavy storms) that causes
piezometric head gradient high enough to overcome the slope stability
Application: typically a hilly landscape with thick soil cover with
unchanneled valleys where soil accumulation and discharge (by
landsliding) alternates cyclically.
Limitation: NOT suitable for deep seated landslides, rocky outcrops, areas
with deep groundwater tables, unstable glacial or postglacial terrains
GIS in Landslide assessment (advanced)
SHALSTAB (SHAllow Landslide STABility)
Theory:
Infinite slope model
Assumptions:
no losses in water balance: effective precipitation equals the rainfall (no evapotranspiration taken into
account), no deep drains and no superficial (overland) flow, only subsurface runoff
runoff trajectories parallel with the slope and slip surface, with the laminar flow (Darcy’s law)
geo-mechanic parameters:
C - cohesive strength of the soil = 0
(no cohesion and no root system reinforcement effect)
φ - internal friction angle = 45°
γ - volume weight ranges from 16-20 kN/m3
Stability model
solve by h/z:
GIS in Landslide assessment (advanced)
SHALSTAB (SHAllow Landslide STABility)
Hydrologic model (transmissivity T vs. rainfall q trough Darcy’s law)
T/q [m]
q/T [1/m]
log (q/T) [1/m]
3162
0.00040
-3.4
1259
0.00079
-3.1
631
0.00158
-2.8
316
0.00316
-2.5
158
0.00633
-2.2
79
0.01266
-1.9
SHALSTAB: solving combined equations of stability and subsurface flow
GIS in Landslide assessment (advanced)
SHALSTAB (SHAllow Landslide STABility)
Training and calibrating
Effects of parametrization
Volume weight and friction angle constant, (allowing C=0 and comparisons between different
landscapes)
Field measurements (area of the sliding body, width at the crown or toe, local slope angle)
Effects of slope angle and drainage area calculation
Minor differences due to slope algorithm type (8 neighboring cells)
Slope angle gradient vs. slip surface angle gradient
Maximum fall vs. multiple direction algorithm for drainage area
Effects of grid size
Since coarser resolution gives smoother slopes coarser grids lack in detailedness
GIS in Landslide assessment (advanced)
SHALSTAB (SHAllow Landslide STABility)
Testing (using field data to accept/reject parametric free model)
Mapping the landslide scar sites and overlaying over SHALSTAB model
Comparing different scenarios
GIS in Landslide assessment (advanced)
SINMAP
Concept: similarly as SHALSTAB couple the slope stability and
hydrologic model but trough the concept of stability index/safety
factor (SI/FS) also emphasizing topographic influence (in a way
SHALSTAB is a special case of SINMAP)
As SHALSTAB considers cases of pore water pressure increase due to heavy
rainstorm events
Also holds true for hilly landscape with unchanneled valleys
Involves probabilistic uncertainty in parameter setting (such as cohesion, bulk
density and so forth)
Faces the same limitations as SHALSTAB (terrain types, high dependence on
DEM accuracy and accuracy of landslide inventory)
GIS in Landslide assessment (advanced)
SINMAP
Theory
Infinite slope model (with perpendicular dimensioning)
Factor of safety (suppressing vs. driving forces)
Assumptions
As in SHALSTAB apart from cohesion dimensionless factor
GIS in Landslide assessment (advanced)
SINMAP
Theory
Hydrologic model - Topographic Wetness Index (TWI)
Specific catchment area a=A/b
based on the approach of hollow areas
(topographic convergence areas)
Assuming that:
Subsurface flow follows topographic gradient
(superficial topography is used for calculation of a)
Recharge R (heavy rainfall, snowmelt) = lateral discharge q
Flux of the recharge = Transmissivity T *sinθ
(T=kuniform *h)
Lateral discharge:
Relative wetness w=hw/h now
with max set to 1 (superficial flow)
R/T becomes a singleparameter that treats climatic and hydrologic influence
GIS in Landslide assessment (advanced)
SINMAP
Theory
Stability model – Stability index
From
to
where r=0,5 but C, R/T and tan φ are normally distributed variables (uncertainty involved)
Spatial and temporal probability is included ranging from worst case scenario
(lowest C, highest R/T, lowest tan φ) to best case scenario (vice versa)
Probabilities of SI
GIS in Landslide assessment (advanced)
SINMAP
Training and calibrating
Pit filling DEM correction
Effect of slope and flow direction from corrected DEM effects
Specific catchment area calculation
GIS in Landslide assessment (advanced)
GEOtop
Analyzes 3D hydrologic flow (lateral and normal) by solving general case of
Richard’s equation
Uses Bishops failure criteria
Takes antecedent conditions of soil moist into account
GIS in Landslide assessment (advanced)
DYLAM
Also for shallow landsliding
Analyzes dynamic data by time vector of rainfall events (unambiguous
temporal probability)
Requires additional geo-mechanical parameters as constant or float values
(the latter provides temporal probability)
Uses simple subsurface flow hydrology
Final output is factor of safety map based on infinite slope modeling, giving
an actual hazard map for the selected time sequence
Couples the GIS environment trough .asc files
GIS IN GEOLOGY
Exercise 4
4.11.2010.
Miloš Marjanović