Annals of GIS,
Journal Year:
2024,
Volume and Issue:
30(1), P. 1 - 14
Published: Jan. 2, 2024
The
Annual
Meeting
of
the
American
Association
Geographers
(AAG)
in
2023
marked
a
five-year
milestone
since
first
Geospatial
Artificial
Intelligence
(GeoAI)
Symposium
was
held
at
AAG
2018.
In
past
five
years,
progress
has
been
made
while
open
questions
remain.
this
context,
we
organized
an
panel
and
invited
panellists
to
discuss
advances
limitations
GeoAI
research.
commended
successes,
such
as
development
spatially
explicit
models,
production
large-scale
geographic
datasets,
use
address
real-world
problems.
also
shared
their
thoughts
on
current
research,
which
were
considered
opportunities
engage
theories
geography,
enhance
model
explainability,
quantify
uncertainty,
improve
generalizability.
This
article
summarizes
presentations
from
provides
after-panel
organizers.
We
hope
that
can
make
these
more
accessible
interested
readers
help
stimulate
new
ideas
for
future
breakthroughs.
Langmuir,
Journal Year:
2025,
Volume and Issue:
41(5), P. 3490 - 3502
Published: Jan. 31, 2025
The
applications
of
machine
learning
(ML)
in
complex
interfacial
interactions
are
hindered
by
the
time-consuming
process
manual
feature
selection
and
model
construction.
An
automated
ML
program
was
implemented
with
four
subsequent
steps:
data
distribution
analysis,
dimensionality
reduction
clustering,
selection,
optimization.
Without
need
intervention,
descriptors
metal
charge
variance
(ΔQCT)
electronegativity
substrate
(χsub)
(δχM)
were
raised
up
good
performance
predicting
electrochemical
reaction
energies
for
both
nitrogen
(NRR)
CO2
(CO2RR)
on
metal-zeolites
MoS2
surfaces.
important
role
tuning
catalytic
reactivity
NRR
CO2RR
highlighted
from
SHAP
analysis.
It
proposed
that
Fe-,
Cr-,
Zn-,
Nb-,
Ta-zeolites
favorable
catalysts
NRR,
while
Ni-zeolite
showed
a
preference
CO2RR.
elongated
bond
N2
or
bent
configuration
shown
V-,
Co-,
Mo-zeolites,
indicating
molecule
could
be
activated
after
adsorption
pathways.
generalizability
automatically
built
is
demonstrated
to
other
systems
such
as
metal-organic
frameworks
SiO2
useful
tool
accelerate
data-driven
exploration
relationship
between
structures
material
properties
without
selection.
Environment International,
Journal Year:
2022,
Volume and Issue:
170, P. 107574 - 107574
Published: Oct. 8, 2022
The
inconstant
climate
change
and
rapid
urbanization
substantially
disturb
the
global
thermal
balance
induce
severe
urban
heat
island
(UHI)
effect,
adversely
impacting
human
development
health.
Existing
literature
has
revealed
UHI
characteristics
driving
factors
at
an
scale,
but
interactions
between
main
of
a
grid
scale
assessment
on
context
zones
remain
unclear.
Therefore,
based
multidimensional
climatic
socio-economic
statistical
datasets,
multi-time
surface
intensity
(SUHI)
was
investigated
in
this
study
to
analyze
how
natural-anthropogenic
drivers
affect
variance
SUHI
vary
their
importance
for
changes
other
interaction
factors.
results
show
that
mean
value
summer
is
higher
than
winter,
daytime
nighttime
seasonal
daily
scale.
SUHIs
different
have
significant
differences.
When
analyzing
drivers'
contributions
with
LightGBM
model
SHAP
algorithm,
we
know
monthly
precipitation
(PREC),
estimated
population
(POP)
pressure
(PRES)
are
three
major
SUHI.
mainly
PREC,
POP
anthropogenic
emission
(AHE),
influence
rules
natural
driversare
mostly
opposite
daytime.
This
highlights
fundamental
role
background
designing
strategies.
Irrigation
or
artificial
rainfall
will
be
effective
mitigate
low
areas,
while
it
more
reduce
AHE
high
areas.
In
where
greening
can
difficult
most
developed
cities,
reducing
AHE,
increasing
per
capita
GDP
controlling
may
also
contribute
alleviating
provides
ideas
developing
responsive
mitigation
policies
realistic
setting.
Land,
Journal Year:
2023,
Volume and Issue:
12(5), P. 1018 - 1018
Published: May 5, 2023
(1)
Background:
The
aim
of
this
paper
was
to
study
landslide
susceptibility
mapping
based
on
interpretable
machine
learning
from
the
perspective
topography
differentiation.
(2)
Methods:
This
selects
three
counties
(Chengkou,
Wushan
and
Wuxi
counties)
in
northeastern
Chongqing,
delineated
as
corrosion
layered
high
middle
mountain
region
(Zone
I),
(Wulong,
Pengshui
Shizhu
southeastern
mountainous
strong
karst
gorges
II),
area.
used
a
Bayesian
optimization
algorithm
optimize
parameters
LightGBM
XGBoost
models
construct
evaluation
for
each
two
regions.
model
with
accuracy
selected
according
indicators
order
establish
mapping.
SHAP
then
explore
formation
mechanisms
different
landforms
both
global
local
perspective.
(3)
Results:
AUC
values
test
set
mode
Zones
I
II
are
0.8525
0.8859,
respectively,
those
0.8214
0.8375,
respectively.
shows
that
has
prediction
regard
landforms.
Under
landform
types,
elevation,
land
use,
incision
depth,
distance
road
average
annual
rainfall
were
common
dominant
factors
contributing
most
decision
making
at
sites;
fault
river
have
degrees
influence
under
types.
(4)
Conclusions:
optimized
LightGBM-SHAP
is
suitable
analysis
types
landscapes,
namely
region,
gorges,
can
be
internal
decision-making
mechanism
levels,
which
makes
results
more
realistic
transparent.
beneficial
selection
index
system
early
prevention
control
hazards,
provide
reference
potential
hazard-prone
areas
research.
International Journal of Geographical Information Science,
Journal Year:
2023,
Volume and Issue:
37(5), P. 963 - 987
Published: March 24, 2023
AbstractImproving
the
interpretability
of
geospatial
artificial
intelligence
(GeoAI)
models
has
become
critically
important
to
open
'black
box'
complex
AI
models,
such
as
deep
learning.
This
paper
compares
popular
saliency
map
generation
techniques
and
their
strengths
weaknesses
in
interpreting
GeoAI
learning
models'
reasoning
behaviors,
particularly
when
applied
analysis
image
processing
tasks.
We
surveyed
two
broad
classes
model
explanation
methods:
perturbation-based
gradient-based
methods.
The
former
identifies
areas,
which
help
machines
make
predictions
by
modifying
a
localized
area
input
image.
latter
evaluates
contribution
every
single
pixel
model's
prediction
results
through
gradient
backpropagation.
In
this
study,
three
algorithms—the
occlusion
method,
integrated
gradients
class
activation
method—are
examined
for
natural
feature
detection
task
using
algorithms'
are
discussed,
consistency
between
model-learned
human-understandable
concepts
object
recognition
is
also
compared.
experiments
used
GeoAI-ready
datasets
demonstrate
generalizability
research
findings.Keywords:
XAIartificial
intelligencedeep
learningvisualizationGeoAI
Disclosure
statementNo
potential
conflict
interest
was
reported
author(s).Data
codes
availability
statementThe
data
that
support
findings
study
available
at
https://github.com/ASUcicilab/explainable-geoai.
Instructions
on
how
use
provided
README
file.Additional
informationFundingThis
work
supported
part
National
Science
Foundation
under
[awards
2120943,
2230034,
1853864].Notes
contributorsChia-Yu
HsuChia-Yu
Hsu
professional
Arizona
State
University.
His
interests
include
intelligence,
computer
vision,
spatiotemporal
analysis,
applications
climate
change
terrain
research.Wenwen
LiWenwen
Li
professor
geographic
information
science
University
(ASU).
Her
cyberinfrastructure,
big
data,
data-
computation-intensive
environmental
social
sciences.
At
ASU,
she
directs
Cyberinfrastructure
Computational
Intelligence
Lab
(http://cici.lab.asu.edu/)
serves
Research
Director
Spatial
Analysis
Center.
