Evaluation of Lidar-derived DEMs through Terrain
Analysis and Field Comparison
Cody P. Gillin, Scott W. Bailey, Kevin J. McGuire, and Stephen P. Prisley
Abstract
Topographic analysis of watershed-scale soil and hydrological
processes using digital elevation models (
DEMs
) is commonplace,
but most studies have used
DEMs
of 10 m resolution or coarser.
Availability of higher-resolution
DEMs
created from light detec-
tion and ranging (lidar) data is increasing but their suitability
for such applications has received little critical evaluation. Two
different 1 m
DEMs
were re-sampled to 3, 5, and 10 m resolu-
tions and used with and without a low-pass smoothing filter
to delineate catchment boundaries and calculate topographic
metrics. Accuracy was assessed through comparison with field
slope measurements and total station surveys.
DEMs
provided
a good estimate of slope values when grid resolution reflected
the field measurement scale. Intermediate scale
DEMs
were most
consistent with land survey techniques in delineating catchment
boundaries. Upslope accumulated area was most sensitive to
grid resolution, with intermediate resolutions producing a range
of
UAA
values useful in soil and groundwater analysis.
Introduction
Topographic analysis using digital elevation models (
DEMs
) has
become routine in soil and hydrologic sciences, and there has
been considerable assessment of the effects of grid resolution on
topographic metrics. Most watershed-scale studies examined
resolutions of 10 m or coarser and tended to use
DEMs
covering
thousands of hectares. For instance, when researchers examined
slope computed from
DEMs
of different resolutions, they ob-
served that coarser
DEMs
generated lower values (e.g., Isaacson
and Ripple, 1990; Jenson, 1991). Quinn
et al.
(1991) compared
topographic wetness index (
TWI
) computed from 12.5 and 50 m
DEMs
and found higher values for the coarser
DEM
. Many other
studies comparing topographic metric values computed from
a range of
DEMs
reported lower slope, larger upslope accumu-
lated areas (
UAAs
), and higher
TWI
values for coarser
DEMs
(e.g.,
Hancock, 2005; Saulnier
et al.
, 1997; Wolock and Price, 1994;
Zhang and Montgomery, 1994). Variation in topographic metric
values computed from
DEMs
of different resolutions is a result of
discretization effects when the size of
DEM
grid cells is altered
(which can affect the algorithm used to compute a topographic
metric) and the loss of terrain detail (smoothing) that occurs
through
DEM
coarsening (Gallant and Hutchinson, 1996).
Examination of soil and hydrologic variability of small
headwater catchments may be enhanced by higher-resolution
DEM
data that has only recently become available through
light detection and ranging (lidar) technology. Lidar-derived
DEMs
have been shown to be more representative of field slope
measurements (Shi
et al.
, 2012) and field-determined eleva-
tions (Vaze
et al.
, 2010) than
DEMs
created using topographic
maps. However, few studies have assessed variation in
topographic metric values extracted from a range of high-reso-
lution (10 m or less) lidar-derived
DEMs
. Sorensen and Seibert
(2007) coarsened a 5 m lidar-derived
DEM
to 10, 25, and 50 m
resolutions and found median
TWI
values increased with
DEM
grid cell size. Vaze
et al
. (2010) noted changes in
DEM
-delin-
eated catchment boundaries across five lidar-derived
DEMs
as
resolution decreased from 1 to 25 m.
While lidar-derived
DEMs
may represent field conditions
better than topographic maps, their accuracy has been shown
to vary depending on land cover class. For example, previous
studies found elevation errors increased under forest canopy
compared with open areas (Hodgson
et al.
, 2005; Reutebuch
et al.
, 2003; Su and Bork, 2006). Greater
DEM
elevation error
associated with forest canopy may be related to a decrease in
the number of lidar ground returns or off-terrain points incor-
rectly classified as ground (Hodgson
et al
., 2005).
Quinn
et al
. (1991) contended that the resolution of
DEMs
used in hydrologic modeling must reflect topographic features
vital to the hydrologic response, suggesting that resolution of
early
DEMs
was too coarse for accurate modeling of some catch-
ments. Two decades later, high-resolution
DEMs
may offer a lev-
el of topographic detail greater than that controlling surface/
near surface flow pathways. For instance, Bailey
et al.
(2014)
found that a 5 m
DEM
resulted in
UAA
and
TWI
values that were
better correlated with soil horizon thickness and groundwater
fluctuations than metrics calculated from a 1 m
DEM
. Gillin
et
al.
(2014) showed that digital mapping of soils based on
DEM
derived topographic metrics was possible with a smoothed
DEM
. To mitigate landscape roughness, a
DEM
may be coarsened
to a lower resolution through resampling or cell aggregation
(e.g., Band and Moore, 1995; Sorensen and Seibert, 2007; Wu
et al.
, 2008) or smoothed through filtering (e.g., Lillesand and
Kiefer, 2000; Walker and Willgoose, 1999). Filtered
DEMs
retain
general topographic trends better than coarsened
DEMs
while
reducing local roughness created by individual cells (Ham-
mer
et al.
, 1995). Although filtering is a common technique for
smoothing
DEMs
, evaluations of topographic metrics computed
from filtered and unfiltered
DEMs
over a range of resolutions
are limited (e.g., Hammer
et al
., 1995).
This study had three principal objectives. First, we com-
pared differences in shape and area of a catchment delineated
from 1 m
DEMs
interpolated from lidar datasets, as well as
DEMs
aggregated from original 1 m resolution to coarser models (3,
5, and 10 m resolutions) and treated with low-pass smoothing
Cody P. Gillin is with Trout Unlimited, Wenatchee, WA,
and formerly with the Department of Forest Resources and
Environmental Conservation, Virginia Tech, Blacksburg, VA.
Scott W. Bailey is with the USDA Forest Service, Northern
Research Station, North Woodstock, NH (
.
Kevin J. McGuire is with the Virginia Water Resources
Research Center and Department of Forest Resources and
Environmental Conservation, Virginia Tech, Blacksburg, VA.
Stephen P. Prisley is with the Department of Forest Resources
and Environmental Conservation, Virginia Tech, Blacksburg, VA.
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 5, May 2015, pp. 387–396.
0099-1112/15/387–396
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.5.387
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
May 2015
387
