Journal of Forecasting,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 24, 2024
ABSTRACT
The
prediction
and
early
warning
of
soybean
futures
prices
have
been
even
more
crucial
for
the
formulation
food‐related
policies
trade
risk
management.
Amid
increasing
geopolitical
conflicts
uncertainty
in
across
countries
recent
years,
there
significant
fluctuations
global
prices,
making
it
necessary
to
investigate
reveal
price
determination
mechanism,
accurately
predict
trends
future
prices.
Therefore,
this
study
proposes
a
comprehensive
interpretable
framework
forecasting.
Specifically,
employs
set
methodologies.
Using
snow
ablation
optimizer
(SAO),
improves
parameters
time
fusion
transformer
(TFT)
model,
an
advanced
predictive
model
based
on
self‐attention
mechanism.
Besides,
addresses
factors
influencing
constructs
effective
features
through
feature
method.
To
explore
volatility
trends,
original
series
are
decomposed
using
variational
mode
decomposition
(VMD).
This
also
enhances
accuracy
predictions
by
introducing
coefficients
trading
volumes
as
predictors.
empirical
findings
suggest
that
VMD‐SAO‐TFT
interpretability,
offering
implications
decision‐makers
achieve
accurate
agricultural
Water,
Год журнала:
2024,
Номер
16(8), С. 1136 - 1136
Опубликована: Апрель 17, 2024
With
global
warming
and
intensified
human
activities,
extreme
convective
precipitation
has
become
one
of
the
most
frequent
natural
disasters.
An
accurate
reliable
assessment
severe
events
can
support
social
stability
economic
development.
In
order
to
investigate
accuracy
enhancement
methods
data
fusion
strategies
for
events,
this
study
is
driven
by
horizontal
reflectance
factor
(ZH)
differential
(ZDR)
dual-polarization
radar.
This
research
work
utilizes
microphysical
information
storms
provided
radar
variables
construct
event
model.
Considering
problems
high
dimensionality
variable
low
computational
efficiency,
proposes
a
echo-data-layering
strategy.
Combined
with
results
mutual
(MI),
constructs
Bayes–Kalman
filter
(KF)
models
(RF,
SVR,
GRU,
LSTM)
events.
Finally,
comparatively
analyzes
evaluation
effectiveness
efficiency
different
models.
The
show
that
data-layering
strategy
able
reduce
dimensions
256
×
34,978
5
2213,
which
greatly
improves
efficiency.
addition,
correlation
coefficient
interval
III–V
calibration
period
increased
0.9,
overall
model
good.
Among
them,
Bayes–KF-LSTM
best
effect,
Bayes–KF-RF
highest
Further,
five
typical
are
selected
validation
in
study.
stratified
dataset
agrees
well
near-surface
precipitation,
model’s
values
close
observed
values.
completely
offered
dual-polarized
ZH
ZDR
assessment,
provides
wide
range
application
possibilities
Research Square (Research Square),
Год журнала:
2024,
Номер
unknown
Опубликована: Июль 24, 2024
Abstract
The
accurate
cumulative
precipitation
forecasts
are
essential
for
monitoring
water
resources
and
natural
disasters.
Thecombination
of
deep
learning
big
data
has
become
a
new
direction
forecasting.
However,
the
currentlarge
models
still
lacking
in-situ
verification.
To
accomplish
this
goal,
forecasting
performance
astate-of-the-art
model
GraphCast
was
evaluated.
Using
from
2393
observation
stations
the1-3
day
period
as
reference,
we
assessed
in
mainland
China
region
1-3
from2020
to
2021,
utilizing
high-resolution
with
0.25◦×0.25◦
grid
spacing
37
layers
parameters.
ofEuropean
Centre
Medium-Range
Weather
Forecasts
(ECMWF)
also
compared.
results
show
that:
(1)
During
the2020-2021
period,
1-day,
2-day,
3-day
forecasts,
Root
Mean
Square
Error
(RMSE)values
were
primarily
between
0.46
9.38
mm/d,
0.44
9.06
respectively.
TheMean
(ME)
values
mainly
−0.595
1.705
(0.01
mm).
(2)
As
forecast
extends,
forecastingcapability
declines.
(3)
In
various
China,GraphCast
demonstrates
higher
predictive
accuracy
than
ECMWF.
(4)
Compared
ECMWF,
demonstrated
thebest
warm-temperate
humid
sub-humid
north
China,
RMSE
being
approximately
12%higher.
Our
study
indicates
that
significant
potential
Revista de Gestão Social e Ambiental,
Год журнала:
2024,
Номер
18(6), С. e08267 - e08267
Опубликована: Авг. 1, 2024
Objective:
The
objective
of
the
research
was
to
analyze
and
compare
different
machine
learning
models
identify
which
technique
presents
best
performance
in
predicting
hydrometeorological
variables.
Theoretical
Framework:
This
section
main
concepts
that
underpin
work.
Machine
techniques
such
as
support
vector
machines,
decision
trees,
random
forests,
artificial
neural
networks,
gradient
boosting
are
presented,
providing
a
solid
foundation
for
understanding
context
investigation.
Method:
study
uses
comparative
methodology
by
applying
predict
variables
based
on
data
collected
Petrolina-PE.
Various
were
employed
compared.
Data
normalization
performed
through
logarithms,
treatment
included
filling
or
excluding
inconsistent
records.
effectiveness
is
evaluated
using
metrics
Nash-Sutcliffe
efficiency
coefficient,
Willmott
index,
Pearson
correlation
coefficient.
Results
Discussion:
obtained
results
showed
good
predictability,
ranging
from
50
70%
efficiency.
analysis
allowed
identifying
patterns
relationships
between
initial
configurations
algorithms,
contributing
better
processes
their
predictability.
Research
Implications:
By
more
accurate
reliable
forecasts,
presented
can
assist
managers
making
decisions
about
sustainable
use
water
mitigation
natural
disasters
floods.
Originality/Value:
contributes
literature
advancing
estimation
variables,
improving
existing
techniques,
resource
management.
Its
impact
extends
mitigating
risks
associated
with
extreme
hydrological
events
promoting
resources,
sustainability
resilience
aquatic
ecosystems,
essential
face
climate
change
environmental
challenges.
TecnoLógicas,
Год журнала:
2024,
Номер
27(60), С. e3017 - e3017
Опубликована: Июнь 27, 2024
La
emisión
de
gases
efecto
invernadero,
atribuida
directa
o
indirectamente
a
la
actividad
humana,
es
principal
causa
del
cambio
climático
nivel
mundial.
Entre
los
emitidos,
el
dióxido
carbono
(CO2)
que
más
contribuye
variación
espacio
temporal
magnitudes
físicas
como
humedad
relativa,
presión
atmosférica,
temperatura
ambiente
y,
manera
significativa,
precipitación.
El
objetivo
investigación
fue
presentar
un
análisis
predicción
precipitación
mensual
en
departamento
Boyacá
mediante
uso
modelos
basados
aprendizaje
reforzado
(RL,
por
sus
siglas
inglés).
metodología
empleada
consistió
extraer
datos
desde
CHIRPS
2,0
(Climate
Hazards
Group
InfraRed
Precipitation
with
Station
data,
versión
2,0)
con
una
resolución
espacial
0,05°
posteriormente
fueron
preprocesados
para
implementación
enfoques
simulación
Montecarlo
y
profundo
(DRL,
inglés)
proporcionar
predicciones
mensual.
Los
resultados
obtenidos
demostraron
DRL
generan
significativas
Es
esencial
reconocer
convencionales
Aprendizaje
profundo,
Memoria
Corto
Plazo
(LSTM)
Redes
Convolucionales
(ConvLSTM),
pueden
superar
términos
precisión
predicción.
Se
concluye
técnicas
refuerzo
detecta
patrones
información
ser
usados
soporte
estrategias
dirigidas
mitigar
riesgos
económicos
sociales
derivados
fenómenos
climáticos.
ITM Web of Conferences,
Год журнала:
2024,
Номер
65, С. 03007 - 03007
Опубликована: Янв. 1, 2024
Precipitation
expectation
is
a
pivotal
subject
for
the
administration
of
water
assets
and
counteraction
hydrological
calamities.
To
make
precipitation
forecast
find
essential
elements
influencing
precipitation,
this
study
presents
logical
profound
learning
approach
in
two
sections.
The
initial
segment
with
consideration
system
which
could
foresee
while
second
part
clarification
figures
attribution
values
information
weather
conditions
to
evaluate
their
significance.
A
contextual
investigation
led
on
hourly
India’s
population
wise
top
eight
urban
cities.
outcomes
predominantly
demonstrate
that
main
whose
component
esteem
adversely/decidedly
corresponded
its
esteem.
review’s
importance
lies
upgrading
giving
interpretability
through
recognizable
proof
persuasive
variables,
works
long
haul
arranging
more
comprehension
mind-boggling
climate
frameworks.
Decision Analytics Journal,
Год журнала:
2024,
Номер
12, С. 100515 - 100515
Опубликована: Авг. 24, 2024
Rainfall
prediction
significantly
impacts
agriculture,
water
reserves,
and
preparations
for
flooding
conditions.
This
research
examines
the
performance
interpretability
of
machine
learning
(ML)
models
rainfall
in
Republic
Ireland.
The
study
uses
a
brute
force
approach
Leave
One
Feature
Out
(LOFO)
methodology
to
evaluate
model
under
highly
correlated
variables.
Results
reveal
consistent
across
ML
algorithms,
with
average
Area
Under
Curve
Precision-Recall
(AUC-PR)
scores
ranging
from
0.987
1.000,
certain
features
such
as
atmospheric
pressure
soil
moisture
deficits
demonstrating
significant
influence
on
outcomes.SHapley
Additive
exPlanations
(SHAP)
values
provide
insights
into
feature
importance,
reaffirming
significance
prediction.
underscores
importance
selection
enhancing
accuracy
usability
Abstract.
Meteorological
and
hydrological
processes
depend
on
accurate
precipitation
observations.
Most
products
utilize
station-based
observations
directly
or
to
bias
correct
satellite
retrievals.
Thus,
the
validation
of
requires
further
independent
data.
This
study
aims
assess
accuracy
Global
Precipitation
Climatology
Center
(GPCC)
Project
(GPCP)
by
estimating
drought
recovery
time
(DRT)
from
terrestrial
water
storage
anomaly
(TWSA)
acquired
gravimetry
required
amount
across
five
main
Köppen-Geiger
climate
zones.
Station-based
products,
namely
GPCC
Full
Data
Monthly
Product
v2022
GPCP
v3.2
Analysis
Product,
were
utilized
estimate
DRT.
Additionally,
JPL
mascon
G3P
Total
Water
Storage
(TWS)
monthly-solutions
Gravity
Recovery
Climate
Experiment
(GRACE)
GRACE
Follow-On
(GRACE-FO)
missions
also
employed
for
DRT
estimation.
was
estimated
through
following
two
methods:
(1)
deficit,
determined
as
negative
residual
detrended
TWSA
its
climatology,
(2)
amount,
derived
linear
relationship
between
cumulative
smoothed
(cdPA)
TWSA.
The
results
show
no
significant
differences
in
mean
estimations
using
GPCP.
Conversely,
estimation
is
2.6
months
longer
average
than
that
G3P.
equatorial
zone
showed
shortest
estimation,
10.3
months,
while
polar
had
longest,
16.2
months.
Except
zone,
arid
shows
highest
estimations,
13.9
Consistency
methods
high
different
zones,
with
exhibiting
highest,
97.8
%,
lowest,
74.9
%.
Similar
results,
consistency
not
obtained
In
contrast,
approximately
5.0
%
higher
mascon.
findings
based
indicate
a
close
agreement
Moreover,
more
consistent
These
provide
necessary
information
product
characteristics,
which
helps
understanding
meteorological
processes.
Journal of Forecasting,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 24, 2024
ABSTRACT
The
prediction
and
early
warning
of
soybean
futures
prices
have
been
even
more
crucial
for
the
formulation
food‐related
policies
trade
risk
management.
Amid
increasing
geopolitical
conflicts
uncertainty
in
across
countries
recent
years,
there
significant
fluctuations
global
prices,
making
it
necessary
to
investigate
reveal
price
determination
mechanism,
accurately
predict
trends
future
prices.
Therefore,
this
study
proposes
a
comprehensive
interpretable
framework
forecasting.
Specifically,
employs
set
methodologies.
Using
snow
ablation
optimizer
(SAO),
improves
parameters
time
fusion
transformer
(TFT)
model,
an
advanced
predictive
model
based
on
self‐attention
mechanism.
Besides,
addresses
factors
influencing
constructs
effective
features
through
feature
method.
To
explore
volatility
trends,
original
series
are
decomposed
using
variational
mode
decomposition
(VMD).
This
also
enhances
accuracy
predictions
by
introducing
coefficients
trading
volumes
as
predictors.
empirical
findings
suggest
that
VMD‐SAO‐TFT
interpretability,
offering
implications
decision‐makers
achieve
accurate
agricultural
