iScience,
Год журнала:
2024,
Номер
28(1), С. 111681 - 111681
Опубликована: Дек. 25, 2024
The
construction
of
dams
to
intercept
natural
rivers
constitutes
the
most
severe
human
activity
influencing
underlying
surface.
This
study
focuses
on
four
cascade
reservoirs
Lancang
River
and
explores
their
impact
migration
organic
matter
in
sediments.
research
reveals
significant
spatial
variations
total
carbon
(TOC)
nitrogen
concentrations
sediments
reservoirs.
isotopes
indicate
that
terrigenous
is
main
source
TOC
sediments,
contributing
an
average
66.80%.
Endogenous
algal-derived
second
source,
between
14.30%
32.91%.
sources
contributed
from
upstream
are
lowest,
ranging
6.36%
15.33%.
Our
demonstrates
may
significantly
alter
processes
material
river
basin
ecosystem,
particularly
large
reservoir
which
increased
more
endogenous
matter.
Water
security
and
its
sustainable
management
are
critical
to
human
survival
livelihoods,
especially
under
the
dual
pressures
of
climate
change
population
growth.
In
response
these
challenges,
an
increasing
number
natural
watersheds
being
regulated
by
dams
reservoirs,
introducing
significant
complexity
streamflow
modeling.
However,
operation
man-made
infrastructures,
small-scale
ones
managed
local
governments,
is
highly
flexible
irregular,
making
them
difficult
investigate
model
thoroughly.
Remote
sensing
products
can
reveal
reservoir
dynamics
at
larger
spatial
scales,
providing
valuable
data
for
data-scarce
catchments.
This
study
aims
evaluate
a
deep
learning
architecture,
namely
Multi-TimeScale
Long
Short-Term
Memory
(MTS-LSTM),
which
capable
incorporating
multi-source
multi-timescale
simulate
streamflow.
Furthermore,
role
remote
sensing-derived
monthly
storage
anomalies
in
MTS-LSTM
enhancing
daily
reservoir-regulated
simulation
investigated.
The
results
case
on
Yuanjiang
River
Basin
demonstrated
that
effectively
bridge
gap
between
SWAT-simulated
observed
streamflow,
attributed
regulations.
simulated
satisfactory
performance
both
(mean
values
Correlation
Coefficients
[CC]=0.92,
Nash–Sutcliffe
Efficiency
[NSE]=0.81
Kling-Gupta
[KGE]=0.80)
CC=0.79,
NSE=0.58
KGE=0.71)
timescales.
integration
into
has
significantly
enhanced
simulation.
mean
CC,
NSE,
KGE
simulations
showed
improvements
5%,
14%,
respectively.
led
higher
level
accuracy
than
achieved
naive
LSTM
model.
presents
systematic
methodology
enhance
simulations,
with
particular
focus
regions
limited
hybrid
cascade
systems.
