Geocarto International,
Год журнала:
2022,
Номер
37(26), С. 12509 - 12535
Опубликована: Апрель 20, 2022
A
major
earthquake
(6.9
Moment
magnitude)
occurred
in
the
Sikkim
and
Darjeeling
areas
of
Indian
Himalaya
as
well
adjacent
Nepal
on
18th
September
2011,
triggering
a
large
number
landslides.
total
188
landslide
locations
were
extracted
order
to
create
inventory
map
(LIM).
The
earthquake-induced
susceptibility
maps
(LSMs)
created
using
an
Artificial
Neural
Network
(ANN)
model
three
novel
deep
learning
approaches
(DLAs),
namely
Deep
Boosting
(DB),
Learning
(DLNN),
Tree
(DLT),
training
points
22
conditioning
factors.
LSMs
validated
several
statistical
indices
results
showed
optimal
accuracy
for
all
models,
where
DB
yielding
highest
prediction
rate
curve
(PRC)
98.5%.
This
is
followed
by
DLT
(97%),
DLNN
(96%),
ANN
(91%).
demonstrate
maximum
efficacy
proposed
LSM.
Communications Earth & Environment,
Год журнала:
2023,
Номер
4(1)
Опубликована: Май 10, 2023
Abstract
Landslides
are
notoriously
difficult
to
predict
because
numerous
spatially
and
temporally
varying
factors
contribute
slope
stability.
Artificial
neural
networks
(ANN)
have
been
shown
improve
prediction
accuracy
but
largely
uninterpretable.
Here
we
introduce
an
additive
ANN
optimization
framework
assess
landslide
susceptibility,
as
well
dataset
division
outcome
interpretation
techniques.
We
refer
our
approach,
which
features
full
interpretability,
high
accuracy,
generalizability
low
model
complexity,
superposable
network
(SNN)
optimization.
validate
approach
by
training
models
on
inventories
from
three
different
easternmost
Himalaya
regions.
Our
SNN
outperformed
physically-based
statistical
achieved
similar
performance
state-of-the-art
deep
networks.
The
found
the
product
of
precipitation
hillslope
aspect
be
important
primary
contributors
highlights
importance
strong
slope-climate
couplings,
along
with
microclimates,
occurrences.
Geocarto International,
Год журнала:
2021,
Номер
37(23), С. 6713 - 6735
Опубликована: Июль 12, 2021
Flood-susceptibility
mapping
is
an
important
component
of
flood
risk
management
to
control
the
effects
natural
hazards
and
prevention
injury.
We
used
a
remote-sensing
geographic
information
system
(GIS)
platform
machine-learning
model
develop
susceptibility
map
Kangsabati
River
Basin,
India
where
flash
common
due
monsoon
precipitation
with
short
duration
high
intensity.
And
in
this
subtropical
region,
climate
change's
impact
helps
influence
distribution
rainfall
temperature
variation.
tested
three
models-particle
swarm
optimization
(PSO),
artificial
neural
network
(ANN),
deep-leaning
(DLNN)-and
prepared
final
classify
flood-prone
regions
study
area.
Environmental,
topographical,
hydrological,
geological
conditions
were
included
models,
was
selected
based
on
relations
between
potentiality
causative
factors
multi-collinearity
analysis.
The
results
validated
evaluated
using
area
under
receiver
operating
characteristic
(ROC)
curve
(AUC),
which
indicator
current
state
environment
value
>0.95
implies
greater
floods.
AUC
values
for
ANN,
DLNN,
PSO
training
datasets
0.914,
0.920,
0.942,
respectively.
Among
these
showed
best
performance
0.942.
approach
applicable
eastern
part
India,
allow
mitigation
help
improve
region.
Geomatics Natural Hazards and Risk,
Год журнала:
2022,
Номер
13(1), С. 949 - 974
Опубликована: Апрель 11, 2022
Flood
is
a
common
global
natural
hazard,
and
detailed
flood
susceptibility
maps
for
specific
watersheds
are
important
management
measures.
We
compute
the
map
Kaiser
watershed
in
Iran
using
machine
learning
models
such
as
support
vector
(SVM),
Particle
swarm
optimization
(PSO),
genetic
algorithm
(GA)
along
with
ensembles
(PSO-GA
SVM-GA).
The
application
of
assessment
mapping
analyzed,
future
research
suggestions
presented.
model
was
constructed
based
on
fifteen
causatives:
slope,
slope
aspect,
elevation,
plan
curvature,
land
use,
cover,
normalize
differences
vegetation
index
(NDVI),
convergence
(CI),
topographical
wetness
(TWI),
topographic
positioning
Index
(TPI),
drainage
density
(DD),
distance
to
stream,
terrain
ruggedness
(TRI),
surface
texture
(TST),
geology
stream
power
(SPI)
inventory
data
which
later
divided
by
70%
training
30%
validated
model.
output
evaluated
through
sensitivity,
specificity,
accuracy,
precision,
Cohen
Kappa,
F-score,
receiver
operating
curve
(ROC).
evaluation
method
shows
robust
results
from
(0.839),
particle
(0.851),
(0.874),
SVM-GA
(0.886),
PSO-GA
(0.902).
Compared
have
done
some
methods
commonly
used
this
assessment.
A
high-quality,
informative
database
essential
classification
types
that
very
helpful
improve
performances.
performance
ensemble
better
than
model,
yielding
high
degree
accuracy
(AUC-0.902%).
Our
approach,
therefore,
provides
novel
studies
other
watersheds.
Remote Sensing,
Год журнала:
2023,
Номер
15(3), С. 798 - 798
Опубликована: Янв. 31, 2023
Landslide
is
a
natural
disaster
that
seriously
affects
human
life
and
social
development.
In
this
study,
the
characteristics
effectiveness
of
convolutional
neural
network
(CNN)
conventional
machine
learning
(ML)
methods
in
landslide
susceptibility
assessment
(LSA)
are
compared.
Six
ML
used
study
Adaboost,
multilayer
perceptron
(MLP-NN),
random
forest
(RF),
naive
Bayes,
decision
tree
(DT),
gradient
boosting
(GBDT).
First,
basic
knowledge
structures
CNN
methods,
steps
LSA
introduced.
Then,
11
conditioning
factors
three
categories
Hongxi
River
Basin,
Pingwu
County,
Mianyang
City,
Sichuan
Province
chosen
to
build
train,
validation,
test
samples.
The
models
constructed
based
on
these
For
comparison,
indicator
statistical
maps
(LSMs)
used.
result
shows
can
obtain
highest
accuracy
(86.41%)
AUC
(0.9249)
LSA.
represented
by
mean
variance
TP
TN
perform
more
firmly
possibility
occurrence.
Furthermore,
LSMs
show
all
successfully
identify
most
points,
but
for
areas
with
low
frequency
landslides,
some
insufficient.
model
demonstrates
better
results
recognition
landslides’
cluster
region,
also
related
convolution
operation
takes
surrounding
environment
information
into
account.
higher
concentrative
great
significance
prevention
mitigation,
which
help
efficient
use
material
resources.
Although
performs
than
other
there
still
limitations,
identification
low-cluster
landside
be
enhanced
improving
model.
Land,
Год журнала:
2023,
Номер
12(2), С. 468 - 468
Опубликована: Фев. 13, 2023
Landslides
are
prevalent
in
the
Western
Ghats,
and
incidences
that
happened
2021
Koottickal
area
of
Kottayam
district
(Western
Ghats)
resulted
loss
10
lives.
The
objectives
this
study
to
assess
landslide
susceptibility
high-range
local
self-governments
(LSGs)
using
analytical
hierarchy
process
(AHP)
fuzzy-AHP
(F-AHP)
models
compare
performance
existing
susceptible
maps.
This
never
witnessed
any
massive
landslides
dimension,
which
warrants
necessity
relooking
into
landslide-susceptible
models.
For
AHP
F-AHP
modeling,
ten
conditioning
factors
were
selected:
slope,
soil
texture,
land
use/land
cover
(LULC),
geomorphology,
road
buffer,
lithology,
satellite
image-derived
indices
such
as
normalized
difference
index
(NDRLI),
water
(NDWI),
burn
ratio
(NBR),
soil-adjusted
vegetation
(SAVI).
zones
categorized
three:
low,
moderate,
high.
validation
maps
created
receiver
operating
characteristic
(ROC)
technique
ascertained
performances
AHP,
F-AHP,
TISSA
excellent,
with
an
under
ROC
curve
(AUC)
value
above
0.80,
NCESS
map
acceptable,
AUC
0.70.
Though
is
negligible,
prepared
model
has
better
(AUC
=
0.889)
than
0.872),
0.867),
0.789)
employing
other
matrices
accuracy,
mean
absolute
error
(MAE),
root
square
(RMSE)
also
confirmed
(0.869,
0.226,
0.122,
respectively)
performance,
followed
by
(0.856,
0.243,
0.147,
respectively),
(0.855,
0.249,
0.159,
(0.770,
0.309,
0.177,
most
landslide-inducing
identified
through
LULC,
NDRLI.
Koottickal,
Poonjar-Thekkekara,
Moonnilavu,
Thalanad,
Koruthodu
LSGs
highly
landslides.
identification
areas
diversified
techniques
will
aid
decision-makers
identifying
critical
infrastructure
at
risk
alternate
routes
for
emergency
evacuation
people
safer
terrain
during
exigency.