Revista de Teledetección,
Год журнала:
2024,
Номер
64, С. 49 - 60
Опубликована: Июль 29, 2024
Entre
Ríos
presenta
un
paisaje
particular
con
numerosos
ambientes
contrastantes.
Cartografiar
tanto
los
naturales
y
como
antrópicos
es
una
tarea
frecuente
gracias
a
la
utilización
de
tecnologías
percepción
remota
junto
sistemas
información
geográfica.
Conocer
qué,
cuánto
dónde
se
encuentran
indispensable
para
diseñar
estrategias
uso
sostenible
conservación
recursos
en
territorio.
La
libre
accesibilidad
datos
capacidad
procesamiento
nube
toda
esta
determinante
procesar
clasificar
vegetación
área
determinada.
El
objetivo
fue
confeccionar
mapa
actualizado
rápidamente
actualizable
el
futuro
mismo
método
más
representativos
provincia
conociendo
cuál
mejor
época
del
año
cual
maximiza
porcentaje
acierto
clasificación
algoritmos
automáticos
cada
ambiente.
Utilizar
aprendizajes
útil
conocer
extensión
ecosistemas
amplio
Las
herramientas
Google
Earth
Engine
permitieron
seleccionar
disminuye
probabilidad
error
bajo
costo
computacional
operacional.
Los
resultados
obtenidos
son
indispensables
planificar
políticas
públicas
forma
precisa
certera
las
actividades
productivas,
así
también
naturales.
International Journal of Wildland Fire,
Год журнала:
2023,
Номер
33(1)
Опубликована: Дек. 18, 2023
Background
Extreme
wildfires
pose
a
serious
threat
to
forest
vegetation
and
human
life
because
they
spread
more
rapidly
are
intense
than
conventional
wildfires.
Detecting
extreme
is
challenging
due
their
visual
similarities
traditional
fires,
existing
models
primarily
detect
the
presence
or
absence
of
fires
without
focusing
on
distinguishing
providing
warnings.
Aims
To
test
system
for
real
time
detection
four
Methods
We
proposed
novel
lightweight
model,
called
LEF-YOLO,
based
YOLOv5
framework.
make
model
lightweight,
we
introduce
bottleneck
structure
MobileNetv3
use
depthwise
separable
convolution
instead
convolution.
improve
model’s
accuracy,
apply
multiscale
feature
fusion
strategy
Coordinate
Attention
Spatial
Pyramid
Pooling-Fast
block
enhance
extraction.
Key
results
The
LEF-YOLO
outperformed
comparison
wildfire
dataset
constructed,
with
our
having
excellent
performance
2.7
GFLOPs,
61
FPS
87.9%
mAP.
Conclusions
speed
accuracy
can
be
utilised
real-time
in
fire
scenes.
Implications
facilitate
control
decision-making
foster
intersection
between
science
computer
science.
Forest Ecology and Management,
Год журнала:
2024,
Номер
555, С. 121729 - 121729
Опубликована: Янв. 31, 2024
Accurately
assessing
forest
fire
susceptibility
(FFS)
in
the
Similipal
Tiger
Reserve
(STR)
is
essential
for
biodiversity
conservation,
climate
change
mitigation,
and
community
safety.
Most
existing
studies
have
primarily
focused
on
climatic
topographical
factors,
while
this
research
expands
scope
by
employing
a
synergistic
approach
that
integrates
geographical
information
systems
(GIS),
remote
sensing
(RS),
machine
learning
(ML)
methodologies
identifying
fire-prone
areas
STR
their
vulnerability
to
change.
To
achieve
this,
study
employed
comprehensive
dataset
of
forty-four
influencing
including
topographic,
climate-hydrologic,
health,
vegetation
indices,
radar
features,
anthropogenic
interference,
into
ten
ML
models:
neural
net
(nnet),
AdaBag,
Extreme
Gradient
Boosting
(XGBTree),
Machine
(GBM),
Random
Forest
(RF),
its
hybrid
variants
with
differential
evolution
algorithm
(RF-DEA),
Gravitational
Based
Search
(RF-GBS),
Grey
Wolf
Optimization
(RF-GWO),
Particle
Swarm
(RF-PSO),
genetic
(RF-GA).
The
revealed
high
FFS
both
northern
southern
portions
area,
nnet
RF-PSO
models
demonstrating
percentages
12.44%
12.89%,
respectively.
Conversely,
very
low
zones
consistently
displayed
scores
approximately
23.41%
18.57%
models.
robust
mapping
methodology
was
validated
impressive
AUROC
(>0.88)
kappa
coefficient
(>0.62)
across
all
validation
metrics.
Future
(ssp245
ssp585,
2022–2100)
indicated
along
edges
STR,
central
zone
categorized
from
susceptibility.
Boruta
analysis
identified
actual
evapotranspiration
(AET)
relative
humidity
as
key
factors
ignition.
SHAP
evaluation
reinforced
influence
these
FFS,
also
highlighting
significant
role
distance
road,
settlement,
dNBR,
slope,
prediction
accuracy.
These
results
emphasize
critical
importance
proposed
provide
invaluable
insights
firefighting
teams,
management,
planning,
qualification
strategies
address
future
sustainability.
Environmental Technology & Innovation,
Год журнала:
2024,
Номер
35, С. 103655 - 103655
Опубликована: Май 5, 2024
Forest
fires
pose
a
significant
threat
to
ecosystems
and
socio-economic
activities,
necessitating
the
development
of
accurate
predictive
models
for
effective
management
mitigation.
In
this
study,
we
present
novel
machine
learning
approach
combined
with
Explainable
Artificial
Intelligence
(XAI)
techniques
predict
forest
fire
susceptibility
in
Nainital
district.
Our
innovative
methodology
integrates
several
robust
—
AdaBoost,
Gradient
Boosting
Machine
(GBM),
XGBoost
Random
Deep
Neural
Network
(DNN)
as
meta-model
stacking
framework.
This
not
only
utilises
individual
strengths
these
models,
but
also
improves
overall
prediction
performance
reliability.
By
using
XAI
techniques,
particular
SHAP
(SHapley
Additive
exPlanations)
LIME
(Local
Interpretable
Model-agnostic
Explanations),
improve
interpretability
provide
insights
into
decision-making
processes.
results
show
effectiveness
ensemble
model
categorising
different
zones:
very
low,
moderate,
high
high.
particular,
identified
extensive
areas
susceptibility,
precision,
recall
F1
values
underpinning
their
effectiveness.
These
achieved
ROC
AUC
above
0.90,
performing
exceptionally
well
an
0.94.
The
are
remarkably
inclusion
confidence
intervals
most
important
metrics
all
emphasises
robustness
reliability
supports
practical
use
management.
Through
summary
plots,
analyze
global
variable
importance,
revealing
annual
rainfall
Evapotranspiration
(ET)
key
factors
influencing
susceptibility.
Local
analysis
consistently
highlights
importance
rainfall,
ET,
distance
from
roads
across
models.
study
fills
research
gap
by
providing
comprehensive
interpretable
modelling
that
our
ability
effectively
manage
risk
is
consistent
environmental
protection
sustainable
goals.
Remote Sensing,
Год журнала:
2023,
Номер
15(3), С. 760 - 760
Опубликована: Янв. 28, 2023
Australia
has
suffered
devastating
wildfires
recently,
and
is
predisposed
to
them
due
several
factors,
including
topography,
meteorology,
vegetation,
ignition
sources.
This
study
utilized
a
geographic
information
system
(GIS)
technique
analyze
understand
the
factors
that
regulate
spatial
distribution
of
wildfire
incidents
machine
learning
predict
susceptibility
in
Sydney.
Wildfire
inventory
data
were
constructed
by
combining
fire
perimeter
through
field
surveys
occurrence
gathered
from
visible
infrared
imaging
radiometer
suite
(VIIRS)-Suomi
thermal
anomalies
product
between
2011
2020
for
Sydney
area.
Sixteen
wildfire-related
acquired
assess
potential
based
on
support
vector
regression
(SVR)
various
metaheuristic
approaches
(GWO
PSO)
mapping
In
addition,
2019–2020
“Black
Summer”
acted
as
validation
dataset
predictive
capability
developed
model.
Furthermore,
gain
ratio
(IGR)
method
showed
driving
such
land
use,
forest
type,
slope
degree
have
large
impact
area,
frequency
(FR)
represented
how
influence
occurrence.
Model
evaluation
area
under
curve
(AUC)
root
average
square
error
(RMSE)
used,
outputs
hybrid-based
SVR-PSO
(AUC
=
0.882,
RMSE
0.006)
model
performed
better
than
standalone
SVR
0.837,
0.097)
SVR-GWO
0.873,
0.080)
models.
Thus,
optimizing
with
metaheuristics
improved
accuracy
modeling
The
proposed
framework
can
be
an
alternative
approach
adapted
any
research
related
different
disturbances.
Ecological Indicators,
Год журнала:
2024,
Номер
160, С. 111946 - 111946
Опубликована: Март 1, 2024
Wildfire
risk
prediction
is
a
critical
component
of
disaster
prevention
and
mitigation,
often
closely
associated
with
local
human
activities
in
most
regions.
Recent
studies
demonstrate
that
employing
joint
modeling
techniques
using
diverse
datasets
alongside
Convolutional
Neural
Networks-Long
Short-Term
Memory
Networks
(ConvLSTM)
produces
favorable
predictive
results.
However,
previous
research
inadequately
explored
the
different
impact
factors
across
categories
spatial
orientations,
neglected
fuels
inside
samples.
This
study
focuses
on
six
eastern
provinces
China,
utilizing
multi-source
dataset
comprising
satellite-monitored
wildfire
products
from
2012
to
2022,
along
various
indicating
terrestrial
activities,
simulated
meteorological
elements
high-resolution
vegetation
imagery.
By
introducing
channel
attention
mechanisms
visual
transformer
mode,
this
optimizes
ConvLSTM
model.
Results
indicate
noteworthy
enhancement,
elevating
accuracy,
Kappa
coefficient,
AUC
ROC
curves
91.15%,
80.87%,
97.01%
92.79%,
84.48%,
97.90%,
respectively.
Consequently,
it
reinforces
accuracy
by
increase
structural
features
within
samples
quantifying
differences
importance
factors,
which
also
validated
application
entire
year
2023.
Sensitivity
analysis
reveals
current
model
still
highly
dependent
factors.
Notably,
significantly
surpasses
influence
terrain
ecology
elements,
should
be
considered
further
models.
Thus,
has
developed
methodology
integrating
multiple
sample
features,
could
furnish
high-precision
daily
kilometer-level
products.
method
improve
efficiency
control
improving
narrowing
high-risk
areas.