Remote Sensing,
Год журнала:
2022,
Номер
14(21), С. 5571 - 5571
Опубликована: Ноя. 4, 2022
Soil
texture
is
a
key
soil
property
driving
physical,
chemical,
biological,
and
hydrological
processes
in
soils.
The
rapid
development
of
remote
sensing
techniques
shows
great
potential
for
mapping
properties.
This
study
highlights
the
effectiveness
multitemporal
data
identifying
textural
class
by
using
retrieved
vegetation
properties
as
proxies
impacts
sensors,
modeling
resolutions,
on
accuracy
classification
were
explored.
Multitemporal
Landsat-8
Sentinel-2
images
individually
acquired
at
same
time
periods.
Three
satellite-based
experiments
with
different
inputs,
i.e.,
data,
(excluding
red-edge
parameters),
(including
parameters)
conducted.
Modeling
was
carried
out
three
spatial
resolutions
(10,
30,
60
m)
five
machine-learning
(ML)
methods:
random
forest,
support
vector
machine,
gradient-boosting
decision
tree,
categorical
boosting,
super
learner
that
combined
four
former
classifiers
based
stacking
concept.
In
addition,
novel
SHapley
Addictive
Explanation
(SHAP)
technique
introduced
to
explain
outputs
ML
model.
results
showed
significantly
affected
prediction
accuracy.
models
parameters
performed
consistently
best.
usually
gave
better
fine
(10
medium
(30
than
coarse
(60
resolution.
provided
higher
accuracies
other
highest
values
overall
(0.8429),
kappa
(0.7611),
precision
(0.8378),
recall
rate
(0.8393),
F1-score
(0.8398)
30
m
involving
parameters.
SHAP
quantified
contribution
each
variable
classes,
revealing
critical
roles
separating
loamy
provides
comprehensive
insights
into
effective
various
scales
optical
images.
Geoderma,
Год журнала:
2024,
Номер
447, С. 116912 - 116912
Опубликована: Май 29, 2024
Digital
soil
mapping
relies
on
statistical
relationships
between
profile
observations
and
environmental
covariates
at
the
sample
locations.
However,
inherent
limitations
of
legacy
profiles,
such
as
inaccurate
georeferencing,
could
frequently
introduce
location
errors
into
these
profiles
that
affect
quality
digital
mapping.
To
address
this
challenge,
study
focuses
reducing
error
evaluating
resulting
impact
We
improved
agreement
detailed
descriptive
information
relatively
accurate
(such
elevation,
slope,
land
use)
to
reduce
profiles.
Quantile
regression
forest
models
were
constructed
predict
properties
their
uncertainty
using
before
after
correction.
Our
results
demonstrate
for
majority
variables,
correcting
positional
in
some
extent
enhances
accuracy
The
largest
improvement
was
found
organic
carbon
0–5
cm
depth
interval,
with
21
%
increase
MEC.
reduced
is
particularly
noteworthy
regions
characterized
by
complex
terrain.
In
addition,
details
predicted
maps
errors,
which
better
represent
spatial
variation
across
China.
Besides,
we
also
elevation
primary
controlling
factor
This
research
presents
a
step
towards
producing
high-resolution
high-quality
datasets,
can
provide
essential
support
management
ensure
future
security.
Archives of Agronomy and Soil Science,
Год журнала:
2025,
Номер
71(1), С. 1 - 17
Опубликована: Янв. 6, 2025
Soil
organic
matter
(SOM)
has
a
vital
role
in
maintaining
soil
quality
and
ecosystem
functions.
However,
predicting
its
spatial
distribution
remains
challenging
task
since
it
was
affected
by
various
environmental
covariates.
To
address
this
limitation,
novel
approach
integrating
Bayesian
technique
into
the
random
forest
(RF)
algorithm
proposed
research.
A
total
of
94
surficial
samples
from
top
30
cm
eight
key
covariates
were
utilized
for
training
testing
with
70:30
ratio.
According
to
results,
enhanced
RF
model
demonstrated
significant
improvement
accuracy
(RMSE
=
0.31%;
MAE
0.25%,
R2
0.79,
Acc
0.81)
compared
traditional
0.66%;
0.48%,
0.10,
0.61).
The
four
including
rainfall,
distance
sea,
water
bodies,
altitude
explained
74.07%,
75.37%
variability
SOM
content
models,
respectively.
Locations
high
characterized
abundant
greater
proximity
rivers,
low
elevations.
These
findings
introduce
reliable
context
complex
changes.
International Journal of Remote Sensing,
Год журнала:
2022,
Номер
43(18), С. 6856 - 6880
Опубликована: Сен. 17, 2022
Accurate
mapping
of
soil
organic
carbon
(SOC)
and
inorganic
(SIC)
contents
at
regional
scales
can
be
very
important
for
sustainable
agriculture
management.
Low
variation
in
terrain
attributes
(classically
used
digital
mapping)
low
relief
areas
calls
additional
spatial
data
to
explain
variability.
The
main
objective
this
study
was
evaluate
the
predictive
capability
Sentinel-1
(radar)
Sentinel-2
(optical)
time
series
statistics,
summarized
as
multi-temporal
features
(MTF)
improve
predictions
SOC
SIC
Ghorveh
plain,
located
Kurdistan
Province,
Western
Iran.
A
systematic
grid
sampling
then
employed
collect
150
surface
samples
(0–30
cm)
measurements.
We
applied
boosted
regression
trees
(BRT)
random
forest
(RF)
predict
by
using
covariate
sets
compiled
from
radar
optical
topographic
attributes.
Model
performance,
evaluated
10-fold
cross-validation,
showed
that
RF
set
containing
Sentinel-1,
performed
best
predicting
(RMSE
=
0.23,
ME
0.005,
R2
0.29).
On
other
hand,
SIC,
MTF
ranked
with
BRT
0.77,
ME=
−0.001,
0.48).
indicates
multiple
dates
remote
sensing
results
improved
predictions.
However,
model
performance
moderate
poor,
respectively.
Therefore
more
substantial
studies
would
required
verify
if
computational
effort
is
likely
justified
an
increase
accuracy
general.
