gis-based multivariate statistical analysis for landslide ... filegis-based multivariate statistical...
TRANSCRIPT
GIS-based multivariate statistical analysis for
landslide susceptibility zoning:
a first validation on different areas of Liguria
region (Italy)
R. Marzocchi, A. Rovegno, B. Federici,
R. Bovolenta & R. Berardi
FOSS4G-Europe 2015 July 14th – 17th, Como
DICCA – Polytechnic School – University of Genoa – Italy
Intr
oduction
6.9% of the territory is catalogued as landslide by the IFFI national inventory of landslide (ISPRA , 2013)
Several approaches to assess landslide hazard (Fell et al., 2008): heuristic analysis based on geological and geomorphological criteria and local observations collected by an expert statistical methods usually correlate an inventory of landslides occurred in the past with factors which are supposed to be responsible of slope failure (Cascini, 2008) process-based methods and numerical analysis local analysis
• affecting limited thickness of the loose soil • local contributing factors related to humans (eg. road cuts)
The type of landslides involved in this study are slides and flows (according to the UNESCO WP/WLI, 1993)
associated to intense meteorological events
Intr
oduction
not only emergency management but also prevention
Land use planning supported by
landslide susceptibility zoning
Intr
oduction
The aim Landslide susceptibility zoning on wide area through GIS-based multivariate statistical analysis
To express objectively the landslide susceptibility in accordance with many factors
Software GRASS GIS 7.0
The present research
Criticality analysis of the zoning procedure to develop a guideline for the landslide susceptibility zoning, as an instrument to non-GIS and non-statistical expert users for the choice of factors to be considered.
Predisposing factors to landslide
Logistic multiple regression
Critical analysis
Conclusions
Overview In
troduction
Literature analysis P
redis
posin
g f
acto
rs to landslid
e
1. Slope
2. Land use
3. Lithology
4. Aspect
5. Accumulation
6. Elevation above sea level
7. Distance from road network
8. Climatic aggressivity FFAO
geological Morphological anthropogenic climatic
Lith
olo
gy
stra
tigr
aph
y
Dis
tan
ce fr
om
tec
ton
ic a
lign
men
ts
Soil
typ
e
Soil
dep
th
slo
pe
Mo
rph
olo
gica
l par
amet
ers
asp
ect
Elev
atio
n a
bo
ve s
ea le
vel
Dis
tan
ce fr
om
th
e d
rain
age
net
wo
rk
accu
mu
lati
on
Lan
d u
se
vege
tati
on
Dis
tan
ce fr
om
ro
ad n
etw
ork
Clim
atic
agg
ress
ivit
y FF
AO
Mea
n m
on
thly
rai
n
Natali et al. (2010) x x x
Pradhan & Lee (2010) x x x x x x x x x
Pauditš & Bednárik
(2002) x x x
Lee (2007) (x) x x x x x x x
Dahal et al. (2008) x x x x x x x x x
Cencetti et al. (2010) x x x x x
Dai & Lee (2002) x x (x) x x (x) x
Ayalew & Yamagishi
(2005) x x x x x x x
Raster maps (20x20 m) reclassified for the area that you want to submit a zoning
proven factors of instability
Bivariate analysis
• to reduce number of classes
• to have monotonic (increasing or decreasing) trend of variables
Land use classification
1 Other
2 Buildings
3 Agricultural areas
4 Wood
5 Sparse vegetation
6 Bare soil/rock
𝑃 evento 𝑥 =𝑃 evento⋂𝑥
𝑃 𝑥
Pre
dis
posin
g f
acto
rs to landslid
e
Pre
dis
posin
g f
acto
rs to landslid
e a) FFAO
b) Log(accumulation) c) aspect d) slope e) lithology f) distance from road g) Elevation h) land cover
Bivariate analysis
Generalized linear model (GLM), appropriate for dichotomous data (0 = absence of landslide, 1 = presence of landslide)
𝑷 𝒆𝒗𝒆𝒏𝒕𝒐 𝑿 =𝟏
𝟏 + 𝒆−𝒁
𝒁 = 𝒍𝒐𝒈𝒊𝒕 𝒆𝒗𝒆𝒏𝒕𝒐 = 𝜷𝟎 + 𝜷𝟏 ∙ 𝑿𝟏 + ⋯ + 𝜷𝒑 ∙ 𝑿𝒑
Logis
tic m
ultip
le r
egre
ssio
n
Logistic multiple regression
Logis
tic m
ultip
le r
egre
ssio
n
Logistic multiple regression
1st phase: Calibration
𝑃 𝑒𝑣𝑒𝑛𝑡𝑜 𝑿 =1
1 + 𝑒−𝑍
inventory of landslide (IFFI) associating: P=1 if landslide area P=0 if NOT landslide area
chosen factors
𝑍 = 𝑙𝑜𝑔𝑖𝑡 𝑒𝑣𝑒𝑛𝑡𝑜 = 𝛽0 + 𝛽1 ∙ 𝑋1 + ⋯ + 𝛽𝑝 ∙ 𝑋𝑝
Estimation of regressions coefficients
Logis
tic m
ultip
le r
egre
ssio
n
Logistic multiple regression
2nd phase: Application
𝑃 𝑒𝑣𝑒𝑛𝑡𝑜 𝑿 =1
1 + 𝑒−𝑍
chosen factors
𝑍 = 𝑙𝑜𝑔𝑖𝑡 𝑒𝑣𝑒𝑛𝑡𝑜 = 𝛽0 + 𝛽1 ∙ 𝑋1 + ⋯ + 𝛽𝑝 ∙ 𝑋𝑝
Estimated regressions coefficients
Calculated landslide susceptibility
Critical analy
sis
Critical points of the procedure
• Classification of chosen factors bivariate analysis
• Selection of the indipendent variables of the regression model (factors)
• Choice of the proper area to calibrate the coefficient of the regression model
• Definition of a criterion of classification of the susceptibility values
Critical analy
sis
Test area
Liguria Region (Italy) and its 4 Provinces
Critical analy
sis
Selection of the indipendent factors
2(p
end
enza
e u
sosu
olo
) 3(+
lito
logi
a)
4(+
accu
mu
lazi
on
e)
5(+
esp
osi
zio
ne)
6(+
AC
)
7(+
clas
si q
uo
ta)
8(+
dis
tan
za s
trad
e)
22130000
22135000
22140000
22145000
22150000
22155000
n° regressori
AIC
AIC index (Akaike’s Information Criteria)
Defined the calibration area...
If the adding of a variable has a positive effect on the model, the values of AIC have to decrease
Regione Liguria provincia Imperia provincia Savona provincia Genova provincia La Spezia
0,0000
0,0002
0,0004
0,0006
0,0008
0,0010
0,0012
0,0014
R2
pendenza
uso suolo
litologia
accumulazione
esposizione
quota
distanza strade
aggress.climatica
Critical analy
sis
The proper area to calibrate
R2 correlation coefficient
Defined the factors... indication of the influence factor of the j-th factor on the susceptibility
Slope Land use Lithology Accumulation Aspect Elevation Distance to roads FFAO
Critical analy
sis
Factors vs calibration area
Regression coefficients bj reliability of the factors
Reliable factors • slope • Land use • accumulation • Distance from road
pendenza uso suolo litologia accumulazione esposizione quota distanza strade aggress.climatica
-0,27
-0,22
-0,17
-0,12
-0,07
-0,02
0,03
0,08
0,13
coe
ff. r
egr
ess
ion
e
Regione Liguria
provincia Imperia
provincia Savona
provincia Genova
provincia La Spezia
Slope land use lithology accumul aspect elevation dist road FFAO
Critical analy
sis
Factors vs calibration area
bassa media alta
Importance of spatial variability of factors
Low Medium high
FFAO map
Lithology map
Critical analy
sis
Factors vs calibration area
pendenza uso suolo litologiadist.dastrade
accumul. esposiz. quotaaggress.climatica
Piemonte 0,000136 0,000172 0,007310 0,000054 0,000045 0,001521 0,001026 0,000106
Liguria 0,000659 0,000342 0,000227 0,000196 0,000143 0,000033 0,000032 0,000005
0,0000
0,0010
0,0020
0,0030
0,0040
0,0050
0,0060
0,0070R2
R2 correlation coefficient indication of the influence factor of the j-th factor on the susceptibility
Piemonte
Liguria
Slope land use lithology dist road accumul. aspect elevation FFAO
Critical analy
sis
Classification of the susceptibility values
5.4 e-09
4.3 e-08
Estimated values of susceptibility • Relative results
• Very small values (order of 10-9 - 10-8)
Classification in 3 susceptibility level
low moderate high
Critical analy
sis
Results and validation
signs of instability incipient sliding
Low Moderate High
Critical analy
sis
Results and validation
Low Moderate High
Critical analy
sis
Results and validation
Low Moderate High
First conclusions C
on
clu
sio
ns
• Importance of preliminary analysis of the characteristics of the area to be subjected to zoning bivariate analysis
• Reliable factors: slope, land use, accumulation, distance to roads
• Importance of spatial variability of factors
• Strong influence of the characteristics of the calibration area on zoning results
• Calibration area: not too small, to be representative of all the conditions present in the area to be subjected to zoning, but of more homogeneous characteristics as possible
Future work C
on
clu
sio
ns
• Guidelines for the susceptibility zoning for non-GIS and non-statistical expert users
• Maps with indications of most suitable design solutions according to the characteristics of the critical zones
Co
nclu
sio
ns
This work is financed by the national research program (PRIN 2010-2011) called “Mitigation of landslide risk using sustainable actions” coordinated by University of Salerno.
Thank you for your attention