of understory shrubs but dense midstory beech saplings
and minimal terrain features, and the K sites contained an
intermediate density of vegetation and terrain features. At
each site, the total station was placed directly over a point
established as plot center and leveled to an XY planar accu-
racy of 10 seconds. Data were collected at 3 m intervals along
transects established prior to each survey. Transect lengths
varied depending on visibility through the understory, but
generally ranged from 20 to 35 m. When a terrain feature was
encountered, data points were collected adjacent to and on
the feature, often resulting in greater point densities at these
locations. Vertical (relative elevation), horizontal, and slope
distances, as well as horizontal angle (azimuth in degrees)
were recorded for each point.
Elevation values were extracted from the unfiltered/unfilled
NCALM
and
WMNF
1 m
DEMs
to each survey point. Relative
elevation differences between each benchmark and survey
points indicated by each
DEM
were subtracted from the relative
elevation differences between each benchmark and survey
points indicated by the total station. The absolute value of the
resulting values constitutes overall error. Finally, overall error
was used to calculate Root Mean Squared Error (
RMSE
) as:
RMSE
=
e
n
i
i
n
2
1
=
∑
(1)
where
e
i
are the errors or differences between the elevations
of the total station survey locations and their correspond-
ing
DEM
elevations, and
n
is the number of survey points.
RMSE
offers a common empirical assessment of lidar-derived
DEM
accuracy (Raber
et al.
, 2007). We compared
RMSE
values
between sites, as well as for canopy versus no canopy and ter-
rain feature versus no terrain feature. A decrease in vegetative
cover should correspond with lower
RMSE
caused by higher
lidar ground return densities and fewer off-terrain features
misclassified as ground. Greater
RMSE
for terrain features may
indicate their omission from lidar-derived
DEMs
.
Topographic Metric Computation
DEMs
were used to compute four topographic metrics com-
monly used in soil and hydrologic sciences: slope, planform
curvature, upslope accumulated area (
UAA
), and topo-
graphic wetness index (
TWI
= ln(
α
/tan
β
), where
α
= upslope
accumulated area (
UAA
), and tan
β
= local slope; (Beven and
Kirkby, 1979)). Slope was calculated using the maximum
slope algorithm (Travis
et al
., 1975). Planform curvature was
calculated using the equation described by Zevenbergen
and Thorne (1987).
UAA
is a measure of the amount of area
upslope of a given point on a landscape to which surface flow
is attributable and was calculated using the triangular mul-
tiple flow direction algorithm (Seibert and McGlynn, 2007).
TWI
is a common metric used to simulate surface and shallow
subsurface wetness of different points on a landscape relative
to each other, particularly in hydrologic models (e.g., Niu
et
al.
, 2005; Tague and Band, 2001).
To compare topographic metrics across
DEM
resolution/fil-
ter combinations, we generated random points for topograph-
ic metric value extraction. Point generation was constrained
by a 20 m negative buffer polygon so that all points were also
contained within each watershed boundary. To avoid spatial
autocorrelation, we generated random points at a density
of one point per 1,000 m
2
using the mean area of all
DEM
-
delineated catchments for a total of 421 random points. Slope,
planform curvature,
UAA
, and
TWI
values from each
DEM
resolution/filter combination were extracted to the random
points. Boxplots were used to evaluate differences in topo-
graphic metric distributions across changes in
DEM
resolution
and filtering. Additionally,
UAA
maps were created using the
WMNF DEMs
for unfiltered 1, 3, 5, and 10 m resolutions to dem-
onstrate how
DEM
resolution impacts
UAA
values.
Results
Watershed Boundary Delineation
Watershed boundaries were most similar to the 1950s field-
surveyed boundary when delineated from 3 m and 5 m reso-
lution
DEMs
, regardless of whether a filter was applied (not
shown). The
NCALM
unfiltered 1 m
DEM
excluded a 1 ha region
(approximately 2 percent of the total watershed area) in the
southeastern portion of the watershed that was included in
the field-surveyed boundary, while the
WMNF
unfiltered 10 m
DEM
included a 1 ha area in this same region that was exclud-
ed from the field-surveyed boundary (Figure 2).
Watershed areas were also compared for each
DEM
resolu-
tion/filter combination and with the field-surveyed boundary
of 42.4 ha. Areas ranged from 41.4 ha (unfiltered
NCALM
1 m) to
T
able
2. C
atchment
A
rea
and
M
ean
T
opographic
M
etrics
for
E
ach
dem
R
esolution
/F
ilter
C
ombination
(
lp
denotes
the
A
pplication
of
a
L
ow
-
pass
F
ilter
); M
ean
T
opographic
M
etric
V
alues
are
B
ased
on
a
R
andom
S
ampling
of
10 P
ercent
of
10
m
G
rid
C
ells
that would
F
it
into
E
ach
B
oundary
; A
rea
of
the
F
ield
-
surveyed
B
oundary
is
42.3
ha
Lidar
Dataset
DEM resolution/
filter
Area
(ha)
Slope
(%)
Planform Curvature
(radians/m)
UAA
(m
2
)
TWI
(ln(m
2
))
NCALM
1m
41.4
30.6
2.8
188.2
4.3
NCALM
1mLP
42.3
29.0
0.9
161.3
4.8
NCALM
3m
41.9
28.9
0.4
571.1
6.4
NCALM
3mLP
42.1
28.5
0.1
790.5
6.6
NCALM
5m
42.1
28.8
0.4
1621.1
7.2
NCALM
5mLP
42.0
28.1
0.2
1594.1
7.3
NCALM
10m
42.2
28.4
-0.5
3286.9
8.2
NCALM
10mLP
41.5
27.4
-0.3
3410.8
8.5
WMNF
1m
42.4
30.2
2.4
173.6
4.4
WMNF
1mLP
42.3
28.8
0.9
201.8
4.9
WMNF
3m
42.3
29.3
1.4
858.8
6.4
WMNF
3mLP
41.8
28.4
0.0
937.6
6.6
WMNF
5m
42.3
28.6
-0.5
1668.7
7.2
WMNF
5mLP
42.1
28.0
0.1
1539.5
7.4
WMNF
10m
43.0
28.2
-0.2
2623.1
8.2
WMNF
10mLP
41.6
27.4
-0.6
3097.7
8.5
390
May 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
