Geocarto International,
Journal Year:
2021,
Volume and Issue:
37(25), P. 7462 - 7487
Published: Aug. 31, 2021
This
study
presents
two
machine
learning
models,
namely,
the
light
gradient
boosting
(LightGBM)
and
categorical
(CatBoost),
for
first
time
predicting
flash
flood
susceptibility
(FFS)
in
Wadi
System
(Hurghada,
Egypt).
A
inventory
map
with
445
sites
was
produced
randomly
divided
into
groups
training
(70%)
testing
(30%).
Fourteen
controlling
factors
were
selected
evaluated
their
relative
importance
occurrence
prediction.
The
performance
of
models
assessed
using
various
indexes
comparison
to
common
random
forest
(RF)
method.
results
show
areas
under
receiver
operating
characteristic
curves
(AUROC)
above
97%
all
that
LightGBM
outperforms
other
terms
classification
metrics
processing
time.
developed
FFS
maps
demonstrate
highly
populated
are
most
susceptible
floods.
present
proves
employed
algorithms
(LightGBM
CatBoost)
can
be
efficiently
used
mapping.
Geomatics Natural Hazards and Risk,
Journal Year:
2023,
Volume and Issue:
14(1)
Published: May 4, 2023
This
study
aims
to
examine
three
machine
learning
(ML)
techniques,
namely
random
forest
(RF),
LightGBM,
and
CatBoost
for
flooding
susceptibility
maps
(FSMs)
in
the
Vietnamese
Vu
Gia-Thu
Bon
(VGTB).
The
results
of
ML
are
compared
with
those
rainfall-runoff
model,
different
training
dataset
sizes
utilized
performance
assessment.
Ten
independent
factors
assessed.
An
inventory
map
approximately
850
sites
is
based
on
several
post-flood
surveys.
randomly
split
between
(70%)
testing
(30%).
AUC-ROC
97.9%,
99.5%,
99.5%
CatBoost,
RF,
respectively.
FSMs
developed
by
methods
show
good
agreement
terms
an
extension
flood
inundation
using
model.
models'
showed
10–13%
total
area
be
highly
susceptible
flooding,
consistent
RRI's
map.
that
downstream
areas
(both
urbanized
agricultural)
under
high
very
levels
susceptibility.
Additionally,
input
datasets
tested
determine
least
number
data
points
having
acceptable
reliability.
demonstrate
can
realistically
predict
FSMs,
regardless
samples.
International Journal of Applied Earth Observation and Geoinformation,
Journal Year:
2023,
Volume and Issue:
122, P. 103401 - 103401
Published: July 14, 2023
Flash
floods
are
among
the
world
most
destructive
natural
disasters,
and
developing
optimum
hybrid
Machine
Learning
(ML)
models
for
flash
flood
susceptibility
(FFS)
modeling
remains
a
challenge.
This
study
proposed
novel
intelligence
algorithms
based
on
of
several
ensemble
ML
(i.e.,
Bagged
Flexible
Discriminant
Analysis
(BAFDA),
Extreme
Gradient
Boosting
(XBG),
Rotation
Forest
(ROF)
Boosted
Generalized
Additive
Model
(BGAM))
wrapper-based
factor
optimization
Recursive
Feature
Elimination
(RFE)
Boruta)
to
improve
accuracy
FFS
mapping
at
Neka-Haraz
watershed
in
Iran.
In
addition,
Random
Search
(RS)
method
is
meta-optimization
developed
hyper-parameters.
considers
20
conditioning
factors
(CgFs)
380
non-flood
locations
create
geospatial
database.
The
performance
each
model
was
evaluated
by
area
under
receiver
operating
characteristic
(ROC)
curve
(AUC)
validation
methods,
such
as
efficiency.
demonstrated
good
performance,
with
BGAM-Boruta
achieving
highest
(AUC
=
0.953,
Efficiency
0.910),
followed
ROF-Boruta
0.952),
ROF-RFE
0.951),
BAFDA-Boruta
0.950),
BGAM-RFE
ROF
0.949),
BGAM
0.948),
BAFDA-RFE
0.943),
XGB-Boruta
BAFDA
0.939),
XGB-RFE
0.938)
XGB
0.911).
model,
regional
coverage
about
46%
high
very
areas.
Moreover,
revealed
that
distance
river,
slope,
rainfall,
altitude,
road
CgFs
significant
this
region.
Remote Sensing,
Journal Year:
2024,
Volume and Issue:
16(5), P. 858 - 858
Published: Feb. 29, 2024
Flood
susceptibility
mapping
plays
a
crucial
role
in
flood
risk
assessment
and
management.
Accurate
identification
of
areas
prone
to
flooding
is
essential
for
implementing
effective
mitigation
measures
informing
decision-making
processes.
In
this
regard,
the
present
study
used
high-resolution
remote
sensing
products,
i.e.,
synthetic
aperture
radar
(SAR)
images
inventory
preparation
integrated
four
machine
learning
models
(Random
Forest:
RF,
Classification
Regression
Trees:
CART,
Support
Vector
Machine:
SVM,
Extreme
Gradient
Boosting:
XGBoost)
predict
Metlili
watershed,
Morocco.
Initially,
12
independent
variables
(elevation,
slope
angle,
aspect,
plan
curvature,
topographic
wetness
index,
stream
power
distance
from
streams,
roads,
lithology,
rainfall,
land
use/land
cover,
normalized
vegetation
index)
were
as
conditioning
factors.
The
dataset
was
divided
into
70%
30%
training
validation
purposes
using
popular
library,
scikit-learn
(i.e.,
train_test_split)
Python
programming
language.
Additionally,
area
under
curve
(AUC)
evaluate
performance
models.
accuracy
results
showed
that
XGBoost
predicted
with
AUC
values
0.807,
0.780,
0.756,
0.727,
respectively.
However,
RF
model
performed
better
at
prediction
compared
other
applied.
As
per
model,
22.49%,
16.02%,
12.67%,
18.10%,
31.70%
watershed
are
estimated
being
very
low,
moderate,
high,
highly
susceptible
flooding,
Therefore,
integration
data
could
have
promising
predicting
similar
environments.
