Water,
Год журнала:
2023,
Номер
15(22), С. 3928 - 3928
Опубликована: Ноя. 10, 2023
The
long
short-term
memory
network
(LSTM)
model
alleviates
the
gradient
vanishing
or
exploding
problem
of
recurrent
neural
(RNN)
with
gated
unit
architecture.
It
has
been
applied
to
flood
forecasting
work.
However,
data
have
characteristic
unidirectional
sequence
transmission,
and
architecture
LSTM
establishes
connections
across
different
time
steps
which
may
not
capture
physical
mechanisms
be
easily
interpreted
for
this
kind
data.
Therefore,
paper
investigates
whether
a
positive
impact
is
still
better
than
RNN
in
We
establish
models,
analyze
structural
differences
impacts
two
models
transmitting
data,
compare
their
performance
also
apply
hyperparameter
optimization
attention
mechanism
coupling
techniques
improve
an
optimizing
hyperparameters
using
BOA
(BOA-RNN),
(BOA-LSTM),
MHAM
hidden
layer
(MHAM-RNN),
(MHAM-LSTM)
Bayesian
algorithm
(BOA)
multi-head
(MHAM),
respectively,
further
examine
effects
as
underlying
cross-time
scale
bridging
forecasting.
use
measured
process
LouDe
HuaYuankou
stations
Yellow
River
basin
evaluate
models.
results
show
that
compared
model,
under
1
h
forecast
period
station,
same
structure
improves
four
indicators
Nash–Sutcliffe
efficiency
coefficient
(NSE),
Kling-Gupta
(KGE),
mean
absolute
error
(MAE),
root
square
(RMSE)
by
1.72%,
4.43%,
35.52%
25.34%,
station
significantly.
In
addition,
situations,
outperforms
most
cases.
experimental
suggest
simple
internal
more
suitable
work,
while
methods
such
match
well
propagation
negative
on
accuracy.
Overall,
analyzes
from
multiple
perspectives
provides
reference
subsequent
modeling.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Авг. 15, 2024
Spatial
accurate
mapping
of
land
susceptibility
to
wind
erosion
is
necessary
mitigate
its
destructive
consequences.
In
this
research,
for
the
first
time,
we
developed
a
novel
methodology
based
on
deep
learning
(DL)
and
active
(AL)
models,
their
combination
(e.g.,
recurrent
neural
network
(RNN),
RNN-AL,
gated
units
(GRU),
GRU-AL)
three
interpretation
techniques
synergy
matrix,
SHapley
Additive
exPlanations
(SHAP)
decision
plot,
accumulated
local
effects
(ALE)
plot)
map
global
erosion.
respect,
13
variables
were
explored
as
controlling
factors
erosion,
eight
them
speed,
topsoil
carbon
content,
clay
elevation,
gravel
fragment,
precipitation,
sand
content
soil
moisture)
selected
important
via
Harris
Hawk
Optimization
(HHO)
feature
selection
algorithm.
The
four
models
applied
performance
was
assessed
by
measures
consisting
area
under
receiver
operating
characteristic
(AUROC)
curve,
cumulative
gain
Kolmogorov
Smirnov
(KS)
statistic
plots.
results
revealed
that
GRU-AL
model
considered
most
accurate,
revealing
38.5%,
12.6%,
10.3%,
12.5%
26.1%
lands
are
grouped
at
very
low,
moderate,
high
classes
hazard,
respectively.
Interpretation
interpret
contribution
impact
input
model's
output.
Synergy
plot
exhibited
with
DEM
moisture
predictions.
ALE
showed
precipitation
had
negative
feedback
prediction
Based
SHAP
presented
highest
Results
highlighted
new
regions
latitudes
(southern
Greenland
coast,
hotspots
in
Alaska
Siberia),
which
Ecological Informatics,
Год журнала:
2024,
Номер
82, С. 102769 - 102769
Опубликована: Авг. 11, 2024
Desertification
is
one
of
the
most
significant
environmental
and
social
challenges
globally.
Monitoring
desertification
dynamics
quantitatively
identifying
contributions
its
driving
factors
are
crucial
for
land
restoration
sustainable
development.
This
study
develops
a
standardized
methodological
framework
that
combines
with
mechanisms
at
pixel
level,
applied
to
northern
China
from
2000
2020.
Using
multisource
data
employing
Time
Series
Segmentation
Residual
Trend
analysis
(TSS-RESTREND)
method
alongside
geographical
detector,
we
assessed
reversion,
expansion,
abrupt
change
processes,
along
impacts
interactions
natural
human
were
assessed.
Over
past
two
decades,
proportion
desertified
decreased
by
5.60%.
Notably,
32.88%
area
experienced
while
only
5.86%
underwent
expansion.
Abrupt
changes
in
both
reversed
expanding
areas
observed,
primarily
central
western
regions,
these
concentrated
periods
2009–2011
2014–2016.
The
various
different
sub-regions
exhibited
spatial
heterogeneity.
Increased
precipitation,
temperature,
evapotranspiration
contributed
reversion
area,
wind
speed
influenced
eastern
area.
Additionally,
population
density
afforestation
activities
also
promoted
reversion.
In
contrast,
precipitation
increased
temperature
expansion
areas,
respectively,
exacerbating
this
process.
Overall,
between
enhanced.
Future
control
ecological
engineering
planning
should
focus
on
coupling
effects
relevant
vegetation
changes.
