Land,
Год журнала:
2025,
Номер
14(4), С. 713 - 713
Опубликована: Март 26, 2025
Dunes
are
key
geomorphological
features
in
aeolian
environments,
and
their
automated
mapping
is
essential
for
ecological
management
sandstorm
disaster
early
warning
desert
regions.
However,
the
diversity
complexity
of
dune
morphology
present
significant
challenges
when
using
traditional
classification
methods,
particularly
feature
extraction,
model
parameter
optimization,
large-scale
mapping.
This
study
focuses
on
Gurbantünggüt
Desert
China,
utilizing
Google
Earth
Engine
(GEE)
cloud
platform
alongside
multi-source
remote
sensing
data
from
Landsat-8
(30
m)
Sentinel-2
(10
m).
By
integrating
three
deep
learning
models—DeepLab
v3,
U-Net,
U-Net++—this
research
evaluates
impact
batch
size,
image
resolution,
structure
segmentation
performance,
ultimately
producing
a
high-precision
type
map.
The
results
indicate
that
(1)
size
significantly
affects
optimization.
Increasing
4
to
12
improves
overall
accuracy
(OA)
69.65%
84.34%
89.19%
92.03%
Sentinel-2.
further
16
slower
OA
improvement,
with
reaching
86.63%
92.32%,
suggesting
gradient
optimization
approaches
saturation.
(2)
higher
resolution
greatly
enhances
ability
capture
finer
details,
(OA:
92.45%)
being
5.82%
than
86.63%).
(3)
U-Net
performs
best
images
92.45%,
F1:
90.45%),
improving
by
0.13%
compared
DeepLab
provides
more
accurate
boundary
delineation.
v3
demonstrates
greater
adaptability
low-resolution
images.
presents
approach
integrates
offering
framework
dynamic
monitoring
fine-scale
desert’s
geomorphology.
Journal of Geophysical Research Atmospheres,
Год журнала:
2025,
Номер
130(4)
Опубликована: Фев. 12, 2025
Abstract
Air
quality
in
India
faces
significant
risk
from
agricultural
residue
burning,
especially
Punjab
and
Haryana,
which
are
pivotal
to
the
world's
second‐largest
agrarian
economy.
This
study
quantifies
emissions
post‐monsoon
biomass
burning
(10
October–30
November
2022)
these
states
using
VIIRS
fire
detection
data
Sentinel‐2‐derived
burnt
areas.
Ground
validation
via
district‐level
surveys
aligns
with
findings
of
our
study.
Results
show
51%
total
crop
area
was
burned
(14,700
km
2
Punjab;
8,300
Haryana),
leading
substantial
PM
2.5
(54.28
Gg;
7.94
Gg),
CH
4
(25.63
3.75
CO
(1,100.3
195.7
NH
3
(0.83
0.15
SO
(0.68
0.12
(62.1
11.04
Gg).
Emissions
about
6.5
times
higher
than
Haryana
attributable
greater
(∼14,700
),
yield,
elevated
residue‐to‐crop
ratios.
Compared
VIIRS,
Sentinel‐2
provides
approximately
3.6
emission
estimates,
reflecting
improved
detection.
District‐level
variations
underscore
influence
diverse
farming
practices,
weather,
management.
An
uncertainty
analysis,
derived
multiple
estimates
methodologies,
highlights
regional
disparities:
exhibits
highest
both
CO,
respectively,
showing
least.
Understanding
uncertainties
is
vital
for
forecasting
air
pollution
downwind
cities
such
as
New
Delhi
formulating
targeted
mitigation
strategies.
Remote Sensing,
Год журнала:
2025,
Номер
17(4), С. 722 - 722
Опубликована: Фев. 19, 2025
Tropical
evergreen
forests
represent
the
richest
biodiversity
in
terrestrial
ecosystems,
and
fine
spatial-temporal
resolution
mapping
of
these
is
essential
for
study
conservation
this
vital
natural
resource.
The
current
methods
tropical
frequently
exhibit
coarse
spatial
lengthy
production
cycles.
This
can
be
attributed
to
inherent
challenges
associated
with
monitoring
diverse
surface
changes
persistence
cloudy,
rainy
conditions
tropics.
We
propose
a
novel
approach
automatically
map
annual
10
m
forest
covers
from
2017
2023
Sentinel-2
Dynamic
World
dataset
biodiversity-rich
conservation-sensitive
Central
African
Republic
(CAR).
Copernicus
Global
Land
Cover
Layers
(CGLC)
Forest
Change
(GFC)
products
were
used
first
track
stable
samples.
Then,
initial
cover
maps
generated
by
determining
threshold
each
yearly
median
probability
maps.
From
2023,
modified
finally
produced
NEFI
(Non-Evergreen
Index)
images
estimated
thresholds.
results
proposed
method
achieved
an
overall
accuracy
>94.10%
Cohen’s
Kappa
>87.63%
across
all
years
(F1-Score
>
94.05%),
which
represents
significant
improvement
over
performance
previous
methods,
including
CGLC
based
on
World.
Our
findings
demonstrate
that
provides
detailed
characteristics
time-series
change
Republic,
substantial
consistency
years.
Earth system science data,
Год журнала:
2025,
Номер
17(2), С. 741 - 772
Опубликована: Фев. 26, 2025
Abstract.
The
production
and
evaluation
of
the
analysis-ready
cloud-optimized
(ARCO)
data
cube
for
continental
Europe
(including
Ukraine,
UK,
Türkiye),
derived
from
Landsat
dataset
version
2
(ARD
V2)
produced
by
Global
Land
Analysis
Discovery
(GLAD)
team
covering
period
2000
to
2022,
is
described.
consists
17
TB
at
a
30
m
resolution
includes
bimonthly,
annual,
long-term
spectral
indices
on
various
thematic
topics,
including
surface
reflectance
bands,
normalized
difference
vegetation
index
(NDVI),
soil
adjusted
(SAVI),
fraction
absorbed
photosynthetically
active
radiation
(FAPAR),
snow
(NDSI),
water
(NDWI),
tillage
(NDTI),
minimum
(minNDTI),
bare
(BSF),
number
seasons
(NOS),
crop
duration
ratio
(CDR).
was
developed
with
intention
provide
comprehensive
feature
space
environmental
modeling
mapping.
quality
time
series
assessed
(1)
assessing
accuracy
gap-filled
bimonthly
artificially
created
gaps;
(2)
visual
examination
artifacts
inconsistencies;
(3)
plausibility
checks
ground
survey
data;
(4)
predictive
tests,
examples
organic
carbon
(SOC)
land
cover
(LC)
classification.
reconstruction
demonstrates
high
accuracy,
root
mean
squared
error
(RMSE)
smaller
than
0.05,
R2
higher
0.6,
across
all
bands.
indicates
that
product
complete
consistent,
except
winter
periods
in
northern
latitudes
high-altitude
areas,
where
cloud
density
introduce
significant
gaps
hence
many
remain.
check
further
shows
logically
statistically
capture
processes.
BSF
showed
strong
negative
correlation
(−0.73)
coverage
data,
while
minNDTI
had
moderate
positive
(0.57)
Eurostat
practice
data.
detailed
temporal
characteristics
provided
different
tiers
predictors
this
proved
be
important
both
regression
LC
classification
experiments
based
60
723
LUCAS
observations:
(tier
4)
were
particularly
valuable
mapping
SOC
LC,
coming
out
top
variable
importance
assessment.
Crop-specific
(NOS
CDR)
limited
value
tested
applications,
possibly
due
noise
or
insufficient
quantification
methods.
made
available
https://doi.org/10.5281/zenodo.10776891
(Tian
et
al.,
2024)
under
CC-BY
license
will
continuously
updated.
