ArticlePDF Available

Uncertainty in urban forest canopy assessment: Lessons from Seattle, WA, USA

Authors:

Abstract and Figures

Increasing urbanization around the globe is leading to concern over the loss of tree canopy within cities, but quantifying urban forest canopy cover can be difficult. We discuss methods of assessing canopy cover within cities, and then use a case study of Seattle, WA, USA to examine issues of uncertainty in canopy cover assessment. We find that uncertainty is often not reported, and when reported, may be biased. Based on these findings, we provide a list of recommendations for those undertaking canopy cover assessment in complex urban environments.
Content may be subject to copyright.
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx
Contents
lists
available
at
ScienceDirect
Urban
Forestry
&
Urban
Greening
jo
ur
nal
ho
me
page:
www.elsevier.com/locate/ufug
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA
Jeffrey
J
Richardson,
L.
Monika
Moskal
School
of
Environmental
and
Forest
Sciences,
University
of
Washington,
Seattle,
WA,
USA
a
r
t
i
c
l
e
i
n
f
o
Keywords:
Canopy
cover
Heterogeneity
Object-based
image
analysis
Remote
sensing
Sampling
a
b
s
t
r
a
c
t
Increasing
urbanization
around
the
globe
is
leading
to
concern
over
the
loss
of
tree
canopy
within
cities,
but
quantifying
urban
forest
canopy
cover
can
be
difficult.
We
discuss
methods
of
assessing
canopy
cover
within
cities,
and
then
use
a
case
study
of
Seattle,
WA,
USA
to
examine
issues
of
uncertainty
in
canopy
cover
assessment.
We
find
that
uncertainty
is
often
not
reported,
and
when
reported,
may
be
biased.
Based
on
these
findings,
we
provide
a
list
of
recommendations
for
those
undertaking
canopy
cover
assessment
in
complex
urban
environments.
© 2013 Elsevier GmbH. All rights reserved.
Introduction
The
amount
of
tree
canopy
within
a
city
has
become
a
grow-
ing
concern
in
municipalities
worldwide
as
urbanization
has
led
to
land-use
conversion
and
a
corresponding
loss
of
urban
forests
(Pauchard
et
al.,
2006;
Nowak
et
al.,
2010).
The
benefits
of
tree
canopy
in
mitigating
the
negative
effects
of
air
pollution,
atmospheric
carbon
dioxide,
storm
water
runoff,
and
other
envi-
ronmental
problems
(Xiao
et
al.,
2000;
Brack,
2002)
have
led
to
an
increasing
desire
to
stop
or
reverse
the
losses
in
canopy
cover
to
provide
a
public
benefit.
In
addition,
tree
canopy
has
been
found
to
positively
impact
social
factors
such
as
human
health,
property
val-
ues,
and
well-being
(Ulrich
et
al.,
1991;
Tzoulas
et
al.,
2007;
Wolf,
2007).
Potential
negatives
of
urban
forests
such
as
property
hazard
and
loss
of
garden
space
are
often
overlooked,
but
the
overwhelm-
ing
trend
is
to
associate
tree
canopy
with
a
societal
benefit.
Municipalities
have
sought
to
precisely
quantify
these
benefits
to
understand
the
value
of
their
urban
forests
and
set
management
goals
(McPherson
et
al.,
2011).
While
any
number
of
measures
can
be
used
to
quantify
canopy,
the
simplest
and
most
often
used
is
the
percent
canopy
cover
(CC).
This
is
a
measure
of
the
fractional
projected
area
of
tree
canopy
above
ground-level
expressed
as
a
percentage
ranging
from
0
to
100
(Walton
et
al.,
2008).
It
should
be
noted
that
there
are
issues
with
the
implementation
of
CC
as
a
measure:
the
term
is
often
confused
with
canopy
closure,
there
is
often
confusion
as
to
whether
inter-leaf
gaps
within
a
canopy
should
be
accounted
for
or
ignored,
and
it
is
often
unclear
where
the
threshold
lies
between
tree
canopy
and
the
canopies
of
shrubs
and
other
low
vegetation
(Jennings
et
al.,
1999).
For
example,
an
Corresponding
author.
Tel.:
+1
206
778
4290.
E-mail
address:
jeffjr@uw.edu
(J.J.
Richardson).
investigator
measuring
CC
from
the
ground
will
be
better
able
to
directly
observe
the
tree/shrub
height
threshold
and
observe
or
measure
smaller
gaps
in
the
canopy
(King
and
Locke,
2013)
than
an
investigator
interpreting
an
aerial
photograph
with
0.5
ft
pixel
res-
olution.
Despite
these
issues,
the
CC
measure
is
still
widely
adopted,
and
can
serve
as
a
useful
benchmark
for
assessing
a
municipality’s
quantity
of
urban
forest.
Methodologies
of
canopy
cover
assessment
The
methodologies
most
often
used
to
assess
urban
forest
CC
fall
into
two
groups:
remote
sensing
based
raster
methods
that
pro-
duce
a
census
of
a
city’s
land
cover,
and
sampling-based
methods
that
estimate
city-wide
CC
via
a
subset
of
points
or
plots
within
a
city.
Walton
et
al.
(2008)
provided
a
thorough
overview
of
remote
sensing
techniques,
highlighting
the
various
methods
based
on
aerial
and
satellite
imagery.
In
these
cases,
raster
surfaces
are
used
to
either
classify
a
scene
into
canopy
and
non-canopy
components,
or
used
to
directly
estimate
the
amount
of
CC
within
each
pixel.
Of
note
among
remote
sensing-based
methods
is
the
increasing
use
of
both
object-based
image
analysis
(OBIA)
techniques
and
aerial
LiDAR
(Light
Detection
and
Ranging).
OBIA
is
a
technique
that
breaks
an
aerial
or
satellite
image
into
individual
clusters
of
pixels
(segments)
and
then
classifies
those
segments
based
on
rules
related
to
color,
shape,
or
texture
to
identify
areas
of
canopy
cover.
Aerial
LiDAR
datasets
consist
of
a
three-dimensional
point
cloud
collected
using
an
airplane
or
helicopter
that
can
be
used
to
pro-
duce
rasters
of
CC
and
the
related
leaf
area
index
(LAI)
alone
or
in
conjunction
with
imagery
(Ria˜
no
et
al.,
2004;
Lee
and
Lucas,
2007;
Richardson
et
al.,
2009).
OBIA
and
LiDAR
can
also
be
combined
to
produce
estimates
of
CC
(MacFaden
et
al.,
2012).
Sampling-based
methods
are
different
in
that
they
rely
on
sampling
CC
to
produce
areal
wide-estimates.
They
require
less
1618-8667/$
see
front
matter ©
2013 Elsevier GmbH. All rights reserved.
http://dx.doi.org/10.1016/j.ufug.2013.07.003
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
2J.J.
Richardson,
L.M.
Moskal
/
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx
technology
and
the
individual
samples
are
often
considered
to
be
“true”
since
technicians
are
manually
measuring
CC.
Sampling
is
often
performed
in
the
field
using
fixed
radius
plots
or
based
on
manual
photo
interpretation
of
remotely
sensed
imagery
(USDA
Forest
Service,
2012).
Sampling-based
methods
rely
on
a
sampling
design
and
statistical
inference
to
arrive
at
city-wide
estimates
of
CC.
Because
of
the
perception
that
field
or
photo
interpretation
based
methods
are
more
accurate,
these
methods
are
often
used
to
assess
the
accuracy
of
CC
derived
from
a
remote
sensing
based
raster
methodology.
A
field-based
census
of
all
trees
within
a
city
could
potentially
be
also
used
to
assess
CC,
but
methods
to
obtain
such
a
census,
such
as
those
utilizing
citizen
scientists,
are
only
in
their
infancy
(Urban
Forest
Map,
2012).
No
matter
the
method
used
to
assess
CC
for
a
municipality,
the
estimate
will
contain
some
uncertainty.
One
difficulty
is
in
how
to
quantify,
express,
and
interpret
this
uncertainty
(Atkinson
and
Foody,
2006),
a
set
of
issues
further
confounded
by
the
complex-
ities
facing
municipalities
that
may
have
political
and
economic
pressures
to
achieve
a
certain
level
of
CC.
A
long
history
of
canopy
assessment
in
Seattle,
WA
USA
serves
as
a
case
study
for
exploring
issues
of
uncertainty
in
CC
assessment.
History
of
canopy
cover
assessment
in
Seattle
The
city
of
Seattle
has
a
climate
that
is
well
suited
for
growing
large
trees.
The
area
of
the
city
was
almost
completely
logged
by
the
early
20th
century,
but
contains
many
mature
second-growth
trees
and
an
extensive
network
of
parks
and
green
spaces.
Rapid
urbanization
in
Seattle
and
the
surrounding
region
in
the
last
40
years
has
led
to
land-use
change
and
a
subsequent
loss
of
canopy
(American
Forests,
1999;
Alberti
et
al.,
2004).
The
initial
reports
of
canopy
loss
led
the
City
of
Seattle
to
produce
an
Urban
Forest
Man-
agement
Plan
that
called
for
increasing
the
city
wide
CC
from
18%
in
2004
to
30%
by
2037
(City
of
Seattle,
2007).
The
plan
cited
CC
of
40%
in
Seattle
in
1972,
although
the
accuracy
of
that
estimate
is
unclear.
Several
studies
have
been
performed
since
the
Manage-
ment
Plan
was
published.
The
City
of
Seattle
commissioned
NCDC
Imaging
(no
longer
in
business)
in
2008
to
produce
two
CC
esti-
mates
for
2002
and
2007.
The
company
created
categorical
rasters
of
land
use.
The
canopy
portions
of
these
maps
were
summed
to
produce
CC
estimates
of
22.5%
in
2002
and
22.9%
in
2007
(NCDC
Imaging,
2009).
An
i-Tree
Eco
analysis
was
performed
in
2010
to
quantify
urban
ecosystem
services,
and
also
produced
a
CC
esti-
mate
of
26.3%
(Ciecko
et
al.,
2012).
As
part
of
the
present
paper,
the
authors
created
their
own
CC
estimate
for
Seattle
using
an
OBIA
method,
estimating
CC
at
29.6%.
A
point-based
accuracy
assess-
ment
of
the
OBIA
method
yielded
a
CC
estimate
of
26.3%.
The
i-Tree
Canopy
web
application
(USDA
Forest
Service,
2012),
which
uses
a
point-based
methodology,
was
also
used
by
the
authors
of
this
study
to
produce
a
CC
estimate
of
Seattle
in
2012
of
28.5%.
Lessons
from
Seattle
Table
1
shows
a
large
variation
in
assessed
Seattle
CC
over
time,
with
multiple
values
for
identical
dates
such
as
1972,
2002
and
2009.
While
it
is
possible
that
a
trend
may
exist
within
these
findings,
it
is
difficult
to
draw
conclusions
without
a
measure
of
uncertainty.
American
Forests
Assessments
Assessments
used
to
determine
the
1972
(15%)
and
1996
CC
(10%
and
13%)
values
relied
on
classification
methods
based
on
Landsat
imagery
and
limited
plot-based
sampling
(American
Forests,
1999).
Nowak
and
Greenfield
(2010)
found
that
percent
tree
cover
from
the
2001
National
Land
Cover
Database,
which
also
uses
Landsat
imagery,
significantly
underestimated
CC
in
devel-
oped
lands
by
an
average
of
13.7%
nationally.
The
relatively
coarse
pixel
size
of
Landsat
(30
m)
can
cause
difficulty
in
urban
areas,
where
individual
and
small
clumps
of
trees
dominate
the
canopy
(Nowak
and
Greenfield,
2010)
The
plot-based
method
used
to
esti-
mate
CC
in
1996
is
based
on
7
small
rectangular
plots
within
the
city
manually
interpreted
from
aerial
photos.
While
the
sites
were
selected
to
represent
the
variability
across
the
city,
no
sites
within
parks
were
obtained.
This,
coupled
with
the
low
sample
size
decreases
the
certainty
of
this
estimate
as
a
representation
of
mean
city-wide
CC.
A
CC
estimate
of
40%
was
also
reported
for
1972
in
the
Urban
Forest
Management
Plan
(City
of
Seattle,
2007).
The
source
of
this
value
is
unclear,
but
it
may
be
a
representation
of
the
change
in
the
broader
Seattle
metropolitan
area
captured
in
a
(1998)
American
Forests
report.
If
this
is
the
case,
the
definition
of
the
boundary
of
Seattle
may
be
a
source
of
uncertainty.
A
search
for
publicly
available
GIS
data,
for
example,
provides
at
least
two
different
poly-
gon
boundary
files
for
the
city.
The
city-limits
have
also
changed
over
time
as
neighborhoods
were
annexed
(City
of
Seattle,
2012).
These
boundary-related
sources
of
uncertainty
can
have
a
signifi-
cant
effect
on
the
reported
city-wide
CC
value.
Remote
sensing
raster
based
methods
Two
assessments
for
2002
and
one
assessment
for
2007
relied
on
producing
raster
maps
of
CC
derived
from
remote
sensing
and
other
geospatial
data.
Detailed
methodologies
were
not
published
for
any
of
the
assessments,
nor
were
accuracy
assessments
conducted
(City
of
Seattle,
2007;
NCDC
Imaging,
2009).
It
is
not
uncommon
for
studies
prepared
for
a
non-academic
audience
to
omit
methods
and
accuracy
assessments,
but
it
also
makes
assessing
uncertainty
nearly
impossible.
An
(OBIA)
approach
was
used
in
this
study
to
produce
a
cate-
gorical
raster
map
containing
categories
of
tree,
grass
and
scrub,
bare
ground,
buildings,
and
ground
impervious.
Datasets
used
for
this
classification
included:
2009
NAIP
four
band
imagery,
2003
aerial
LIDAR,
City
of
Seattle
polygon
buildings
layer,
road
polygon
layer,
and
a
polygon
layer
of
major
water
bodies.
By
summing
all
the
tree
pixels
within
a
finished
classification
and
dividing
by
the
total
number
of
pixels
for
Seattle,
a
CC
value
of
29.6%
was
derived.
Fig.
1
shows
a
map
of
CC
for
Seattle.
While
this
result
came
from
a
census
rather
than
a
sample,
biases
are
still
present.
One
way
to
identify
bias
is
by
performing
an
accuracy
assessment.
1000
points
were
randomly
located
within
Seattle
and
assigned
a
class
by
a
trained
photo
interpreter
using
2009
0.5
ft
Aerials
Express
true
color
imagery
and
georeferenced
oblique
angle
aerial
photographs.
The
accuracy
assessment
is
presented
as
an
error
matrix
(Table
2).
This
error
matrix
reveals
several
interesting
statistics
pertinent
to
CC.
First,
it
shows
79.5%
of
the
reference
points
were
correctly
classified
as
trees
in
the
OBIA
map
(Producer’s
Accuracy).
Secondly,
it
shows
that
74.7%
of
the
1000
points
that
were
coincident
with
a
portion
of
the
OBIA
map
classified
as
a
tree
were
reference
trees
and
not
one
of
the
other
classes
(User’s
Accuracy).
Confusion
with
grass
provided
the
largest
source
of
misclassification
of
trees.
Since
producer’s
accuracy
was
greater
than
user’s
accuracy,
more
errors
of
commission
occurred
than
omission
and
thus
the
OBIA
map
was
biased
toward
an
overestimate
of
CC.
Fig.
2
provides
a
different
illustration
of
uncertainty
within
the
classification.
In
Fig.
2,
the
percentage
of
tree
canopy
for
223
0.04
ha
circular
plots
derived
from
the
2009
OBIA
classification
is
compared
to
ocular
estimates
of
canopy
cover
collected
on
the
ground.
This
allows
an
interpretation
of
the
precision
of
the
OBIA
CC
estimates,
as
well
as
bias.
A
linear
regression
shows
that
LULC
explains
69%
of
the
variability
within
the
ocular
estimates
of
CC,
with
a
RMSE
of
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
J.J.
Richardson,
L.M.
Moskal
/
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx 3
Table
1
Estimates
of
city-wide
canopy
cover
for
Seattle.
(RS,
remote
sensing;
GS,
ground
sampling).
Year
Canopy
cover
estimate
Method
Data
source
Citation
1972
15%
Landsat
Sub-pixel
RS,
Landsat
American
Forests
(1999)
1972
40% Unknown
RS,
Landsat
City
of
Seattle
(2007)
1996
10%
Landsat
Sub-pixel
RS,
Landsat
American
Forests
(1999)
1996
10%
Plot-based
photo
interpretation
RS,
Aerial
Photos
American
Forests
(1999)
2002
18%
LiDAR
Analysis
RS,
2000
LiDAR
City
of
Seattle
(2007)
2002
22.5%
Categorical
Raster
Creation
RS,
Unknown
NCDC
Imaging
(2009)
2007
22.9%
Categorical
Raster
Creation
RS,
Unknown
NCDC
Imaging
(2009)
2009
26.4%
Point-based
Random
Sample
RS,
2009
Aerial
Photos
This
Study
2009
29.6% Categorical
Raster
Creation RS,
2003
LiDAR,
2009
Aerial
Photos,
Polygon
Features This
Study
2010
26.3%
i-Tree
Eco
Plots
GS
Ciecko
et
al.
(2012)
2012
28.5%
i-Tree
Canopy
Point-based
Random
Sample
RS,
Google
Maps
This
Study
Fig.
1.
Visualization
of
canopy
cover
pattern
in
Seattle
derived
from
a
categorical
raster
of
land
use/land
cover.
Pixels
are
1
m2.
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
4J.J.
Richardson,
L.M.
Moskal
/
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx
Table
2
Error
matrix
showing
accuracy
for
a
categorical
raster
map
of
land
use
land
cover
in
Seattle.
Overall
accuracy
was
74.00%,
and
the
Khat
was
0.64.
Classification
data
Building
Impervious
Trees
Grass
Water
Bare
ground
Total
Producer’s
accuracy
Reference
data
Building
148
42
17
3
1
1
212
69.81%
Impervious
11 288 18
10
0
0
327
88.07%
Trees
5
29
210
20
0
0
264
79.55%
Grass
2
72
36
73
0
0
183
39.89%
Water
0
0
0
0
7
0
7
100.0%
Bare
Ground
0
5
0
2
0
0
7
0%
Total
166
436
281
108
8
1
1000
User’s
accuracy 89.16% 66.06% 74.73% 67.59% 87.5% 0.00%
Khat
0.86 0.50 0.62 0.60 0.87
0.00
19%.
The
regression
also
suggests
that
the
OBIA
map
overestimates
CC
compared
to
field
measured
CC.
Plot
and
point-based
assessments
Several
assessments
have
relied
on
points
or
plots
to
provide
a
sample
of
CC
within
the
City
of
Seattle,
and
then
have
used
that
sample
to
estimate
CC
for
the
whole
city.
The
2010
i-Tree
Eco
Anal-
ysis
for
Seattle
produced
a
CC
assessment
based
on
223
randomly
selected
plots
stratified
by
city
management
unit
(Ciecko
et
al.,
2012).
One
ocular
estimate
of
CC,
measured
in
categories
of
5%,
was
observed
at
each
0.04
ha
plot
following
the
i-Tree
Eco
protocol,
as
well
as
individual
tree
biophysical
variables
(i-Tree).
We
could
not
find
a
published
source
that
describes
how
the
i-Tree
Eco
soft-
ware
produces
city-wide
estimates
of
CC
given
the
field
sampling
design.
i-Tree
also
produces
a
free
online
tool,
i-Tree
Canopy
that
can
be
used
to
quickly
produce
a
CC
assessment
using
freely
avail-
able
remote
sensing
data
from
Google
Maps
(USDA
Forest
Service,
2011).
We
performed
an
assessment
for
1000
points
generated
by
i-Tree
Canopy
by
visually
classifying
each
point
as
canopy
or
non-
canopy.
The
application
provided
a
CC
estimate
of
28.5%
with
a
standard
error
of
1.4%.
Confidence
intervals
may
be
more
a
more
easily
interpretable
measure
of
uncertainty.
The
95%
confidence
interval
for
the
28.5%
CC
estimate
falls
between
25.7%
and
31.3%.
Fig.
3
shows
the
relationship
between
increasing
sample
size
and
the
decreasing
range
of
the
95%
confidence
interval.
The
accuracy
assessment
conducted
for
the
OBIA
classification
also
provides
a
point-based
measure
of
CC.
Like
the
i-Tree
Canopy
analysis,
1000
points
were
visually
classified
into
canopy
and
non-canopy
using
aerial
imagery.
Instead
of
Google
Map
imagery,
we
utilized
high
Fig.
2.
Comparison
of
ocular
estimates
of
tree
cover
collected
on
the
ground
to
canopy
cover
derived
from
a
categorical
raster
map
of
land
use
land
cover
for
Seattle.
Solid
line
is
the
best
fit
least
squares
regression
and
the
dotted
line
represents
unity.
resolution
aerial
imagery
collected
in
2009
with
a
1
ft
pixel
resolu-
tion,
producing
a
CC
estimate
of
26.3%
(Table
2).
Both
the
i-Tree
Canopy
and
OBIA
accuracy
assessment
point-
based
methods
should
provide
unbiased
estimates
of
CC
assuming
the
points
can
be
classified
without
bias
or
error.
We
have
identified
two
potential
sources
of
error
in
the
classification
of
aerial
photog-
raphy
used
in
this
study:
relief
displacement
and
errors
based
on
interpretation
uncertainty.
Relief
displacement
causes
tall
objects
to
appear
displaced
outward
from
their
true
location.
In
Seattle,
trees
are
often
the
tallest
object
in
a
scene,
and
thus
are
strongly
affected
by
relief
displacement.
The
net
effect
of
canopy
relief
dis-
placement
is
that
more
area
is
covered
by
canopy
in
an
aerial
photograph
than
would
be
expected
by
a
ground
survey.
Since
trees
in
Seattle
often
grown
singly
or
in
small
clusters,
the
effect
of
relief
displacement
is
greater
than
in
larger,
more
continuous
forests
because
more
non-canopy
area
is
obscured
by
the
displaced
tree
canopy.
We
also
noticed
a
difference
in
the
magnitude
of
relief
displacement
between
the
2009
aerial
imagery
and
the
2012
Google
Maps
imagery
through
a
visual
comparison
of
the
two
sets
of
imagery.
Since
relief
displacement
is
more
pronounced
as
the
distance
of
an
object
from
the
location
of
the
camera
increases,
not
all
trees
will
be
affected
by
relief
displacement
to
the
same
degree.
Fig.
4
shows
an
example
of
relief
displacement
for
a
pair
of
tall
trees
in
Seattle
as
seen
in
both
the
2009
and
2012
imagery.
The
greater
relief
displacement
observed
in
the
Google
Maps
imagery
compared
to
the
aerial
imagery
is
a
possible
explanation
for
the
larger
CC
estimate
derived
from
i-Tree
Canopy
(Table
1).
Interpretation
errors
are
a
result
of
the
inability
of
the
analyst
to
classify
a
point
with
complete
certainty.
We
found
that
three
issues
affected
the
ability
of
the
analyst
to
classify
points
with
cer-
tainty:
shadows,
edges,
and
vegetation
height.
Table
3
details
the
129
instances
of
uncertainty
encountered
by
the
analyst
during
the
Fig.
3.
Changes
in
Seattle
city-wide
canopy
cover
estimate
with
the
addition
of
random
points
within
the
city.
Upper
and
lower
95%
confidence
intervals
are
given
by
dotted
lines.
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
J.J.
Richardson,
L.M.
Moskal
/
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx 5
Fig.
4.
Two
images
used
for
classification
showing
relief
displacement
of
tree
canopy.
The
top
image
is
from
Google
Maps
used
by
iTree
Canopy.
The
bottom
image
was
collected
by
Aerials
Express
for
King
County
in
2009.
i-Tree
Canopy
analysis.
Table
3
also
shows
if
the
analyst
chose
to
classify
the
point
as
canopy
or
non-canopy,
in
order
to
help
identify
any
biases
in
the
analysis.
Uncertainty
due
to
height
was
caused
by
the
inability
of
the
analyst
to
determine
if
the
vegetation
met
a
cer-
tain
height
threshold
to
be
considered
tree
canopy.
An
analyst
can
use
shadows
and
other
contextual
information
to
judge
the
height
of
a
tree,
shrub,
or
low
scrub
such
as
brambles,
but
in
many
situ-
ations
the
height
is
not
clear.
In
this
study,
we
chose
an
arbitrary
height
threshold
of
3
m
to
separate
trees
from
other
low
vegeta-
tion.
Uncertainty
due
to
edge
was
a
result
of
the
point
falling
on
or
near
the
boundary
between
canopy
and
non-canopy.
In
a
heteroge-
neous
urban
forest,
these
edges
are
very
common.
Lastly,
shadowed
areas
in
the
image
are
difficult
to
classify.
Large
trees
and
buildings
cast
large
shadows,
making
the
boundary
area
between
canopy
and
non-canopy
classes
in
shadowed
points
difficult
to
ascertain.
Over-
all,
65
uncertain
points
were
classified
as
canopy
and
64
uncertain
points
as
non-canopy.
This
suggests
that
interpretation
errors
did
not
strongly
bias
the
i-Tree
Canopy
CC
results.
The
sub
categories
of
uncertainty
attribution
do
show
that
the
analyst
tended
to
clas-
sify
uncertain
points
in
shadow
and
on
the
edge
as
non-canopy
and
points
of
uncertain
height
as
canopy.
The
analyst
strongly
favored
Table
3
Categorization
of
all
photo-interpreted
points
not
classified
with
100%
certainty.
Classification
Canopy
Non-canopy
Attribution
of
uncertainty
Height
19
13
Edge
15
19
Shadow
6
10
Edge/shadow
8
12
Edge/height
7
6
Height/shadow
1
2
Edge/height/shadow
0
1
Interior
forest
canopy
9
1
Totals
65
64
classifying
points
that
fell
in
shadowed
areas
within
dense
forests
(Interior
Forest
Canopy)
as
canopy.
Conclusions
and
recommendations
We
have
demonstrated
some
of
the
complexities
of
assessing
CC
in
municipal
environments.
The
case
study
of
Seattle
highlights
that
uncertainties
exist
in
all
CC
estimates
independent
of
methodology
used,
and
that
it
can
be
difficult
to
quantify
the
level
of
uncertainty.
This
presents
a
potential
problem
for
those
interested
in
accurate
estimates
of
CC,
especially
when
policy
decisions
and/or
funding
are
tied
to
the
level
of
CC
in
a
city.
In
the
case
of
Seattle,
no
assessment
to
date
has
produced
a
CC
assessment
with
a
clear
quantification
of
uncertainty:
The
raster
based
American
Forests
estimates
do
not
contain
accuracy
assessments
and
the
plot-based
estimate
is
limited
by
small,
unrepresentative
samples.
The
image
interpreted
point-based
estimates
are
biased
by
relief
displacement
and
also
subject
to
uncertainty
in
the
interpretation
of
more
than
10%
of
points
(Table
3).
Categorical
raster
based
measures
either
lack
accu-
racy
assessments
or
rely
on
accuracy
assessments
derived
from
uncertain
point-based
assessments.
Point-based
estimates
are
the-
oretically
well
suited
to
providing
unbiased
estimates
of
CC
with
quantifiable
uncertainty
through
confidence
intervals,
but
points
must
be
classified
accurately,
which
can
be
difficult.
We
present
the
following
recommendations
as
a
guide
to
deci-
sion
makers
when
faced
with
embarking
upon
or
interpreting
data
related
to
municipal
CC:
Acknowledge
that
all
CC
estimates
contain
uncertainty.
If
the
uncertainty
cannot
be
reported
quantitatively,
provide
a
quali-
tative
description
of
potential
sources
of
uncertainty.
Inspect
the
geography
of
the
boundary
used
to
determine
the
extent
of
the
municipality.
Ensure
that
the
boundary
is
consistent
if
multiple
CC
estimates
are
to
be
compared.
When
comparing
CC
estimates
for
different
years,
compare
uncertainties.
If
uncertainties
are
large
or
unknown
for
any
of
the
dates
in
question,
observed
differences
may
not
be
significant.
Please
cite
this
article
in
press
as:
Richardson,
J.J.,
Moskal,
L.M.,
Uncertainty
in
urban
forest
canopy
assessment:
Lessons
from
Seattle,
WA,
USA.
Urban
Forestry
&
Urban
Greening
(2013),
http://dx.doi.org/10.1016/j.ufug.2013.07.003
ARTICLE IN PRESS
G
Model
UFUG-25368;
No.
of
Pages
6
6J.J.
Richardson,
L.M.
Moskal
/
Urban
Forestry
&
Urban
Greening
xxx (2013) xxx–
xxx
See
Nowak
and
Greenfield
(2012)
for
a
broader
discussion
of
quantifying
canopy
cover
change.
If
field
collected
data
point
or
plot
data
were
used,
ensure
that
a
large
number
of
plots
were
collected.
If
a
raster
of
CC
was
created,
ensure
than
an
accuracy
estimate
was
performed.
Record
the
methodology
used
to
conduct
the
CC
assessment.
This
should
be
sufficiently
detailed
that
an
independent
investigator
can
reproduce
the
assessment.
A
random
sample
of
points
presents
an
unbiased
method
for
assessing
CC,
but
may
require
large
numbers
of
points
to
achieve
a
high
degree
of
certainty.
Assessing
points
using
aerial
imagery
may
impart
biases
toward
higher
CC
due
to
relief
displacement.
Visually
assess
imagery
to
determine
the
severity
of
relief
displacement.
Future
research
could
be
directed
toward
developing
the
quantifiable
measures
of
relief
displacement,
and
methods
of
correction.
It
can
be
difficult
and
imprecise
to
differentiate
between
trees,
shrubs,
and
grass
and
low
vegetation
from
aerial
imagery
due
to
the
difficulty
in
assessing
heights.
Consider
using
a
canopy
height
model
derived
from
aerial
LiDAR
to
aid
in
this
classi-
fication.
Alternatively,
uncertain
points
can
be
visited
in
the
field.
Aerial
imagery
will
be
interpreted
with
greater
certainty
if
it
can
be
collected
so
shadows
are
minimized
and
resolution
is
high.
Aerial
LiDAR
can
be
of
aid
if
available
and
collected
at
the
same
time
as
the
imagery.
Future
research
can
help
to
refine
methodologies
of
assessing
CC
in
heterogeneous
urban
environments.
Improvement
in
remote
sensing
technologies,
such
as
the
continuing
development
of
Google
Earth
can
improve
the
quality
of
imagery
publicly
avail-
able,
thereby
making
accuracy
assessments
derived
from
manual
interpretation
more
accurate.
Methods
should
also
be
developed
to
eliminate
or
correct
relief
displacement
from
aerial
and
satellite
imagery.
Lastly,
improvements
in
partnerships
between
munic-
ipalities
and
academic
institutions
can
increase
the
rigor
of
CC
assessments.
Acknowledgements
The
authors
would
like
to
acknowledge
the
contribution
of
com-
ments
from
Jana
Dilley,
Oliver
Bazinet,
Michael
Hannam,
Weston
Brinkley,
and
the
Green
Cities
Research
Alliance
Team.
Funding
and
support
provided
by
the
USDA
Forest
Service
PNW
Research
Station
and
the
American
Recovery
and
Reinvestment
Act.
References
Alberti,
M.,
Weeks,
R.,
Coe,
S.,
2004.
Urban
land-cover
change
analysis
in
Central
Puget
Sound.
Photogramm.
Eng.
Remote
Sens.
70,
1043–1052.
American
Forests,
1998.
Regional
ecosystem
analysis
Puget
Sound
Metropolitan
Area.
American
Forests,
1999.
Urban
ecosystem
analysis
of
Seattle,
Washington.
Atkinson,
P.M.,
Foody,
G.M.,
2006.
Uncertainty
in
remote
sensing
and
GIS:
funda-
mentals.
In:
Atkinson,
P.M.,
Foody,
G.M.
(Eds.),
Uncertainty
in
Remote
Sensing
and
GIS.
John
Wiley
&
Sons,
Ltd.,
Chichester,
UK,
pp.
1–18.
Brack,
C.L.,
2002.
Pollution
mitigation
and
carbon
sequestration
by
an
urban
forest.
Environ.
Pollut.
116,
S195–S200.
Ciecko,
L.,
Tenneson,
K.,
Dilley,
J.,
Wolf,
K.L.,
2012.
Seattle’s
forest
ecosystem
values:
analysis
of
the
structure,
function,
and
economic
benefits.
Green
Cities
Research
Alliance.
City
of
Seattle,
2007.
Urban
forest
management
plan.
City
of
Seattle,
2012.
Seattle
annexation
list,
Retrieved
from
http://clerk.seattle.gov/public/annexations/annex list.htm (retrieved
18.07.12).
i-Tree,
2013.
i-Tree
Eco
user’s
manual.
Jennings,
S.B.,
Brown,
N.D.,
Sheil,
D.,
1999.
Assessing
forest
canopies
and
understorey
illumination:
canopy
closure,
canopy
cover
and
other
measures.
Forestry
72,
59–73.
King,
K.L.,
Locke,
D.H.,
2013.
A
comparison
of
three
methods
for
measuring
local
urban
tree
canopy
cover.
Arboric.
Urban
For.
39,
62–67.
Lee,
A.C.,
Lucas,
R.M.,
2007.
A
lidar-derived
canopy
density
model
for
tree
stem
and
crown
mapping
in
Australian
forests.
Remote
Sens.
Environ.
111,
493–518.
MacFaden,
S.W.,
O’Neil-Dunne,
J.P.M.,
Royar,
A.R.,
Lu,
J.W.T.,
Rundle,
A.G.,
2012.
High-
resolution
tree
canopy
mapping
for
New
York
City
using
lidar
and
object-based
image
analysis.
J.
Appl.
Remote
Sens.
6,
063567.
McPherson,
E.G.,
Simpson,
J.R.,
Xiao,
Q.F.,
Wu,
C.X.,
2011.
Million
trees
Los
Angeles
canopy
cover
and
benefit
assessment.
Landscape
Urban
Plann.
99,
40–50.
NCDC
Imaging,
2009.
Looking
back
and
moving
forward.
City
of
Seattle.
Nowak,
D.J.,
Greenfield,
E.J.,
2010.
Evaluating
the
national
land
cover
database
tree
canopy
and
impervious
cover
estimates
across
the
conterminous
United
States:
a
comparison
with
photo-interpreted
estimates.
Environ.
Manage.
46,
378–390.
Nowak,
D.J.,
Greenfield,
E.J.,
2012.
Tree
and
impervious
cover
change
in
U.S.
cities.
Urban
For.
Urban
Greening
11,
21–30.
Nowak,
D.J.,
Stein,
S.M.,
Randler,
P.B.,
Greenfield,
E.J.,
Comas,
S.J.,
Carr,
M.A.,
Alig,
R.J.,
2010.
Sustaining
America’s
urban
trees
and
forests:
a
forests
on
the
edge
report.
Pauchard,
A.,
Aguayo,
M.,
Pena,
E.,
Urrutia,
R.,
2006.
Multiple
effects
of
urbaniza-
tion
on
the
biodiversity
of
developing
countries:
the
case
of
a
fast-growing
metropolitan
area
(Concepcion,
Chile).
Biol.
Conserv.
127,
272–281.
Ria˜
no,
D.,
Valladares,
F.,
Condes,
S.,
Chuvieco,
E.,
2004.
Estimation
of
leaf
area
index
and
covered
ground
from
airborne
laser
scanner
(lidar)
in
two
contrasting
forests.
Agric.
For.
Meteorol.
124,
269–275.
Richardson,
J.J.,
Moskal,
L.M.,
Kim,
S.-H.,
2009.
Modeling
approaches
to
estimate
effective
leaf
area
index
from
aerial
discrete-return
lidar.
Agric.
For.
Meteorol.
149,
1152–1160.
Tzoulas,
K.,
Korpela,
K.,
Venn,
S.,
Yli-Pelkonen,
V.,
Kazmierczak,
A.,
Niemela,
J.,
James,
P.,
2007.
Promoting
ecosystem
and
human
health
in
urban
areas
using
green
infrastructure:
a
literature
review.
Landscape
Urban
Plann.
81,
167–178.
Ulrich,
R.S.,
Simons,
R.F.,
Losito,
B.D.,
Fiorito,
E.,
Miles,
M.A.,
Zelson,
M.,
1991.
Stress
recovery
during
exposure
to
natural
and
urban
environments.
J.
Environ.
Psychol.
11,
201–230.
2012.
Urban
Forest
Map,
Retrieved
from
http://urbanforestmap.org/
(retrieved
08.06.12).
USDA
Forest
Service,
2011.
i-tree
canopy
technical
notes,
Retrieved
from
http://www.itreetools.org/canopy/resources/iTree
Canopy
Methodology.pdf
(retrieved
11.02.13).
USDA
Forest
Service,
2012.
i-tree
canopy,
Retrieved
from
http://www.itreetools.org/
canopy/index.php
(retrieved
08.06.12).
Walton,
J.T.,
Nowak,
D.J.,
Greenfield,
E.J.,
2008.
Assessing
urban
forest
canopy
cover
using
airborne
or
satellite
imagery.
Arboric.
Urban
For.
34,
334–340.
Wolf,
K.L.,
2007.
City
trees
and
property
values.
Arborist
News
16,
34–36.
Xiao,
Q.F.,
McPherson,
E.G.,
Ustin,
S.L.,
Grismer,
M.E.,
Simpson,
J.R.,
2000.
Winter
rain-
fall
interception
by
two
mature
open-grown
trees
in
Davis,
California.
Hydrol.
Processes
14,
763–784.
... Recently, the increasing accessibility of technologies led to a growing number of high spatial resolution (<3 m) experiments in urban contexts [7,29,32]. For example, studies on urban forest canopy managed to identify its extension [34,35] and tree species [21]. UGI assessment and monitoring differs by regional environment, since each contains different structures of vegetation and complex surfaces [36]. ...
... For example, a study assessing urban vegetation biomass classifies urban vegetation through an object-oriented classification method, using remote sensing images and LiDAR data [11]. Object-based image analysis classifies segments of aerial or satellite images based on rules related to color, shape, and texture [35]. Zhong et al. evaluated the effects of urbanization by calculating two greennessrelated spectral indexes-NDVI and the Enhanced Vegetation Index (EVI)-from Landsat images [12]. ...
... The research presented here visually tests and selects NDVI values greater than 0.18, in line with the values indicated in Table 1 [35], and combines them with height data from LiDAR. Specifically, green areas have been classified as low vegetation (0.00-0.40 m), mainly representing pervious surfaces, medium vegetation (0.40-2.00 m), mainly representing bushes and farmlands, and high vegetation (above 2.00 m). ...