Forestry An International Journal of Forest Research,
Journal Year:
2025,
Volume and Issue:
unknown
Published: Feb. 12, 2025
Abstract
Tree
species
information
is
crucial
both
for
understanding
forest
composition
and
supporting
sustainable
management,
but
also
monitoring
biodiversity
assessing
ecosystem
services.
Remote
sensing
data
has
been
widely
used
to
map
the
spatial
distribution
of
tree
across
large
areas.
However,
there
currently
a
lack
studies
demonstrating
potential
airborne
laser
scanning
collected
during
different
seasons
identify
multiple
individual
species/genera,
including
dead
individuals.
The
main
objective
this
study
was
ecologically
valuable
area
constituting
Polish
part
Białowieża
Forest
using
leaf-on
leaf-off
(ALS)
color-infrared
imagery.
Eleven
living
species/genera
(alder,
ash,
aspen,
birch,
hornbeam,
lime,
maple,
oak,
pine,
spruce
other
deciduous)
four
classes
(dead
deciduous,
snag)
were
classified
at
level.
Applying
Random
Forests
algorithm
set
30
predictor
variables,
15
with
an
overall
accuracy
82
per
cent.
mapping
nearly
20
million
trees
revealed
that
in
2015,
most
common
upper
stands
(20.1
cent),
followed
by
alder
(19.0
cent)
pine
(18.1
cent).
Among
trees,
deciduous
(2.2
(1.7
common.
Our
results
can
serve
as
first
cornerstone
carrying
out
further
in-depth
analyses
remote
exceptional
area.
Remote Sensing of Environment,
Journal Year:
2024,
Volume and Issue:
313, P. 114337 - 114337
Published: Aug. 8, 2024
Information
on
tree
species
across
various
spatial
scales,
from
an
individual
to
a
forest
stand
and
the
broader
landscape,
contributes
accurate
thorough
understanding
of
conditions
either
as
characteristic
or
input
species-dependent
models.
However,
recognition
is
one
most
challenging
tasks
in
remote
sensing
studies,
due
complexity
compositions
canopy
structures
forests,
e.g.,
both
cross-species
similarities
intra-species
variations
commonly
exist
spectral-,
texture-,
structure-
domains.
Over
past
two
decades,
interest
using
close-range
for
has
been
rapidly
growing.
Recent
research
highlighted
needs
further
develop
methods
elevate
their
performance
comparison
with
established
approaches,
address
new
questions
arising
resolutions,
data
coverages,
viewing
geometries,
other
characteristics.
This
work
provides
overview
state-of-the-art
data.
The
summarizes
works
decade,
reviews
state
research,
discusses
prominent
challenges,
investigates
impact
factors,
gaps,
potentials.
Specifically,
sources,
features
derived
each
type
data,
methodologies
applied,
targeted
are
reviewed
detail.
Relevant
machine
learning
(ML)
approaches
grouped
into
conventional
ML
deep-learning
(DL)
categories.
In
category,
reported
studies/results
respect
spectral,
spatial,
temporal
domains
used
sensor
platform.
Despite
significant
efforts
field,
issues
automation,
reliability,
robustness
algorithms
have
only
partially
resolved.
crucial
elements
algorithm
design
what
this
found
worth
careful
consideration
include
types
conditions,
seasonal
variability
phenology,
characteristics
corresponding
feature
selections,
methodology.
Future
studies
recommended
focus
fusion
multi-source
including
passive
active
multispectral
integrate
spectral
structural
information,
use
time-series
enhance
role
phenological
variances
recognition,
development
unsupervised
DL
techniques
improve
accuracy
efficiency.
It
also
promote
sharing
open
standards
facilitate
international
cooperation
communication.
IEEE Transactions on Geoscience and Remote Sensing,
Journal Year:
2024,
Volume and Issue:
62, P. 1 - 19
Published: Jan. 1, 2024
Efficiently
collecting
3D
information
in
complex
forest
environments
holds
significant
importance
for
surveys.
Terrestrial
laser
scanning
or
ground-based
mobile
can
acquire
relatively
smooth
terrain
conditions,
but
they
still
face
the
challenge
of
time-consuming
and
labor
consumption.
In
comparison,
small
unmanned
aerial
vehicle
(UAV)
measurement
systems
with
efficient
autonomous
exploration
capabilities
are
becoming
a
new
close-range
sensing
platform
acquiring
data
within
intricate
canopy
due
to
their
high-efficiency
maneuverability
flexibility.
However,
achieving
UAVs
remains
environments.
this
paper,
we
propose
novel
method
without
relying
on
global
navigation
satellite
system.
Three
heuristic
waypoint
generation
algorithms
introduced
facilitate
unknown
using
quadrotor.
Subsequently,
employing
nonlinear
optimization,
B-spline
curves
employed
generating
smooth,
collision-free,
dynamically
feasible
local
planning
trajectories.
Finally,
sliding
window
strategy
is
utilized
quickly
adjust
trajectory
when
obstacles
like
tree
branches
detected.
This
ensures
that
quadrotor
fly
collisions
by
promptly
re-planning
its
path.
We
have
conducted
series
benchmarked
experiments
plantation
natural
Compared
classic
next-best-view
(NBVP)
method,
proposed
complete
3-7
times
faster
than
NBVP.
fast
UAV
(FUEL)
planner
(FAEP)
methods,
reduces
average
flight
time
distance
60%.
Furthermore,
also
validated
effectiveness
methods
real
experiments.
Forest Ecosystems,
Journal Year:
2024,
Volume and Issue:
11, P. 100238 - 100238
Published: Jan. 1, 2024
Terrestrial
laser
scanning
(TLS)
accurately
captures
tree
structural
information
and
provides
prerequisites
for
tree-scale
estimations
of
forest
biophysical
attributes.
Quantifying
attributes
from
TLS
point
clouds
requires
segmentation,
yet
the
occlusion
effects
severely
affect
accuracy
automated
individual
segmentation.
In
this
study,
we
proposed
a
novel
method
using
ellipsoid
directional
searching
compensation
algorithms
to
alleviate
effects.
Firstly,
region
growing
are
used
determine
location
roots.
Secondly,
neighbor
points
extracted
within
an
neighborhood
mitigate
compared
with
k-nearest
(KNN).
Thirdly,
uniformly
subsampled
by
algorithm
based
on
Fibonacci
principle
in
multiple
spatial
directions
reduce
memory
consumption.
Finally,
graph
describing
connectivity
between
its
neighbors
is
constructed,
it
utilized
complete
segmentation
shortest
path
algorithm.
The
was
evaluated
public
dataset
comprising
six
plots
three
complexity
categories
Evo,
Finland,
reached
highest
mean
77.5%,
higher
than
previous
studies
detection.
We
also
validated
structure
manual
reference
values.
RMSE,
RMSE%,
bias,
bias%
height,
crown
base
projection
area,
surface
volume
were
evaluate
accuracy,
respectively.
Overall,
avoids
many
inherent
limitations
current
methods
can
map
canopy
structures
occluded
complex
stands.
IEEE Transactions on Geoscience and Remote Sensing,
Journal Year:
2024,
Volume and Issue:
62, P. 1 - 21
Published: Jan. 1, 2024
Simultaneous
Localization
and
Mapping
(SLAM)
based
on
Laser
Scanning
(LS)
has
been
quickly
developed
in
the
last
decades.
However,
application
of
LS-SLAM
forest
environments
is
still
at
an
early
development
stage,
limited
by
challenges
posed
environments,
such
as
geometric
degeneration
Global
Navigation
Satellite
System
(GNSS)
denied.
The
applicability,
strengths,
weaknesses
state-of-the-art
methods
not
investigated
even
assessed.
This
study
quantitatively
evaluated
nine
twelve
subtropical
plots
with
different
levels
complexity,
i.e.,
"Easy",
"Medium",
"Difficult".
In
addition,
a
robust
3D
method
specially
designed
for
real-time
mapping
proposed.
solution
extracts
angle-based
features
applies
continuous
filter
2D
angle
image
space
to
identify
stable
enhancing
data
alignment
performance.
benchmarking
results
indicated
that
1)
LiDAR-only
SLAM
presented
average
trajectory
accuracy
15
cm
25
easy
medium
plots,
respectively,
failed
difficult
plots.
2)
LiDAR-IMU
equivalent
or
better
comparison
methods,
yet
their
performance
was
3)
back-end
optimization
significantly
improved
localization
results,
10
level
succeeded
all
4)
proposed
5
complexity
categories.
Compared
estimation
doubled,
RMSE
decreasing
from
level.
5)
inertial
measurement
unit
(IMU)
data,
properly
feature
extraction
method,
loop
closure,
modules
play
crucial
role
resolving
aggressive
motion
robustness.
These
outcomes
provide
valuable
guidelines
researchers
practitioners
select
existing
solutions
conditions
applications,
while,
opening
discussion
future
developments
are
necessary.
