PE&RS May 2015 - page 391

43.0 ha (unfiltered WMNF 10 m) (Table 2). The mean of water-
sheds delineated using NCALM DEMs was 42.0 ha and the mean
of watersheds delineated using WMNF DEMs was 42.2 ha. In
general, coarse (10 m) and fine (1 m) DEMs tended to generate
watersheds containing areas least similar to the overall means,
while intermediate (3 m and 5 m) DEMs tended to generate
watersheds containing areas most similar to the overall means
(Table 2) and the conventionally surveyed boundary.
Comparison of Field and DEM Slope Measurements
The 1 m DEMs generated slope values most different from
field measurements, with the exception of the steepest slopes
(Figure 3). A one-way analysis of variance indicated no
statistically significant differences (α = 0.05) between DEMs of
the same resolution/filter combination derived from NCALM
and WMNF lidar datasets, so for ease of visualization only
NCALM scatterplot data are shown. Twenty-nine percent of
the NCALM 1 m DEM slopes and 23 percent of the WMNF 1 m
DEM slopes exhibited a difference from field slope greater
than 10 percent. In general, difference between field and DEM
slope values decreased with DEM coarsening. Filtering made
no significant impact on DEM slope computation. A one-way
analysis of variance comparing field slope measurements with
DEM-computed slopes indicated no statistically significant dif-
ferences between the means (α = 0.05).
Total Station Ground Surveys
To determine if small scale terrain features were included
when lidar returns were classified as ground points and used
to interpolate a bare earth model, we compared RMSE for loca-
tions with and without terrain features. We also compared
RMSE for the cleared rain gage area with RMSE for sites under
mature forest canopy to determine the effects of canopy on
lidar in a northern hardwood forest. In the rain gage clearing,
the RMSE was 1.06 m for the NCALM DEM and 1.24 m for the
WMNF DEM. Under canopy, the RMSE was 1.32 m for the NCALM
DEM and 1.52 m for the WMNF DEM (Table 3). Locations on or
adjacent to terrain features exhibited an RMSE of 1.49 m for
the NCALM DEM and 1.66 m for the WMNF DEM, while locations
without terrain features exhibited an RMSE of 1.03 m for the
NCALM DEM and 1.18 m for the WMNF DEM (Table 3). RG5 had
the greatest RMSE value when computed by survey site (Table
3). Greater RMSE values were observed for both datasets at lo-
cations with terrain features versus non-feature survey points.
In most cases, total station measurements yielded a greater
relative elevation difference than the DEM between benchmark
and terrain feature survey locations.
Topographic Metric Comparison
Boxplots were used to compare the distributions of topo-
graphic metric values computed for each DEM resolution/filter
combination (Figure 4). A one-way analysis of variance indi-
cated no statistically significant differences (α = 0.05) between
DEMs of the same resolution/filter combination derived from
NCALM and WMNF lidar datasets, so for ease of visualization
only WMNF distributions are shown. Filtering and DEM coars-
ening decreased slope variance and interquartile range (IQR).
Slope IQR decreased from 17 percent to 7 percent (Figure 4a).
Median planform curvature did not change with filtering or
cell aggregation (Figure 4b). Planform curvature variance was
consistently lower for the filtered version of each DEM resolu-
tion (Levene’s test statistic = 47.22, α = 0.05). The 3 m and 5
m DEMs tended to produce planform curvature values with
the narrowest distributions (Figure 4b). Median UAA values
increased for both the NCALM and WMNF DEMs from 188 m2
and 174 m2 to 3411 m2 and 3098 m2, respectively. UAA maps
indicated that for finer-resolution DEMs the watershed was
dominated by grid cells with small UAAs (Plate 1). As DEMs
were coarsened, UAA values became larger, and median TWI
T able 3. RMSE for Each Total S tation Survey Site, Locations Under Mature
C anopy and in the Rain Gage C learing, and L ocations With and Without T er-
rain Features. Relative Elevation Differences between T otal Station Survey
Benchmark/Survey Locations and Corresponding dem B enchmark/Survey
L ocations were Used to Compute rmse
Site ID NCALM WMNF
LowerK 0.69 0.73
UpperK 0.73 0.71
SO2 0.77 0.98
RG5 1.84 2.07
Canopy 1.32 1.52
Clearing 1.06 1.24
Feature 1.49 1.66
No Feature 1.03 1.18
Figure 3. Field slope values measured using a clinometer for 75
locations corresponding with soil pits and shallow subsurface
wells were compared with the difference between field and dem
slope measurements computed using the maximum slope algo-
rithm for ncalm 1, 3, 5, and 10 m dems.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2015 391