MethodsX,
Год журнала:
2023,
Номер
10, С. 102034 - 102034
Опубликована: Янв. 1, 2023
Machine
Learning
models
have
become
a
fruitful
tool
in
water
resources
modelling.
However,
it
requires
significant
amount
of
datasets
for
training
and
validation,
which
poses
challenges
the
analysis
data
scarce
environments,
particularly
poorly
monitored
basins.
In
such
scenarios,
using
Virtual
Sample
Generation
(VSG)
method
is
valuable
to
overcome
this
challenge
developing
ML
models.
The
main
aim
manuscript
introduce
novel
VSG
based
on
multivariate
distribution
Gaussian
Copula
called
MVD-VSG
whereby
appropriate
virtual
combinations
groundwater
quality
parameters
can
be
generated
train
Deep
Neural
Network
(DNN)
predicting
Entropy
Weighted
Water
Quality
Index
(EWQI)
aquifers
even
with
small
datasets.
original
was
validated
its
initial
application
sufficient
observed
collected
from
two
aquifers.
validation
results
showed
that
only
20
samples,
provided
enough
accuracy
predict
EWQI
an
NSE
0.87.
However
companion
publication
Method
paper
El
Bilali
et
al.
[1].
•Development
generate
environment.•Training
deep
neural
network
quality.•Validation
sensitivity
analysis.
Ecological Indicators,
Год журнала:
2023,
Номер
154, С. 110863 - 110863
Опубликована: Сен. 2, 2023
Monitoring
the
spatial
distribution
and
sources
of
heavy
metals
(HM)
in
soil
is
essential
for
avoiding
health
risks
achieving
sustainable
utilization.
Multiple
geospatial
data,
including
remote
sensing,
climate,
topography
were
used
to
extract
environmental
covariates.
Additionally,
scene
was
employed
as
alternative
data
land
use/land
cover
describe
urban
functions
human
activity
intensity
more
detail.
After
converting
a
uniform
resolution
30
m,
these
covariates
adopted
characterize
several
common
HM
soil,
copper
(Cu),
chromium
(Cr),
lead
(Pb),
nickel
(Ni),
zinc
(Zn).
The
RReliefF
algorithm
identify
important
variables.
quantification
models
established
using
back-propagation
neural
network
(BPNN)
deep
(DNN).
Besides,
impact
distance
from
scenes
on
analyzed.
result
demonstrated
that
key
covariate
estimating
soil.
Compared
with
BPNN,
DNN
model
provided
better
accuracy
(R2
=
0.67–0.75)
estimation
five
elements.
Therefore,
map
concentrations
at
grid
scale
m.
highest
risk
pollution
are
industrial
areas,
residential
road,
commercial
concentration
negatively
correlated
scenes.
effective
distances
areas
about
2000
road
500
Sustainability,
Год журнала:
2023,
Номер
15(17), С. 13043 - 13043
Опубликована: Авг. 29, 2023
As
particulate
organic
carbon
(POC)
from
lakes
plays
an
important
role
in
lake
ecosystem
sustainability
and
cycle,
the
estimation
of
its
concentration
using
satellite
remote
sensing
is
great
interest.
However,
high
complexity
variability
water
composition
pose
major
challenges
to
algorithm
POC
Class
II
water.
This
study
aimed
formulate
a
machine-learning
predict
compare
their
modeling
performance.
A
Convolutional
Neural
Network–Long
Short-Term
Memory
(CNN–LSTM)
based
on
spectral
time
sequences
was
proposed
construct
model
Sentinel
2
images
surface
sample
data
Chaohu
Lake
China.
comparison,
performances
Backpropagation
Network
(BP),
Generalized
Regression
(GRNN),
(CNN)
models
were
evaluated
for
inversion
concentration.
The
results
show
that
CNN–LSTM
obtained
higher
prediction
precision
than
BP,
GRNN,
CNN
models,
with
coefficient
determination
(R2)
0.88,
root
mean
square
error
(RMSE)
3.66,
residual
deviation
(RPD)
3.03,
which
are
6.02%,
22.13%,
28.4%
better
model,
respectively.
indicates
effectively
combines
spatial
temporal
information,
quickly
captures
time-series
features,
strengthens
learning
ability
multi-scale
conducive
improving
offers
good
support
source
monitoring
assessment
lakes.
