Point Cloud Volume

1
Accurate Retrievals of Forest Structure and
Biomass from Echidna Ground-based,
Upward-Scanning, Full-Waveform Lidar
T. Yao,1 X. Yang,1 F. Zhao,1 A. Strahler,1 C.
Woodcock,1C. Schaaf,1 D. Jupp,2 D. Culvenor,3 G.
Newnham,3 J. Lovell,3 W. Ni-Meister4
1Boston University 2CSIRO Marine and Atmospheric
Research 3CSIRO Sustainable Ecosystems 4Hunter
College of CUNY
Overview Our research uses a ground-based,
scanning lidar instrument to retrieve forest
canopy structural information, including stand
height, mean tree diameter, basal area, stem
count density, woody biomass, leaf area index,
and foliage profile, and links this information
to airborne and spaceborne lidars to provide
large-area mapping of structural and biomass
parameters. The terrestrial lidar instrument,
Echidna, developed by CSIRO Australia, allows
rapid acquisition of vegetation structure data
that can be readily integrated with
downward-looking airborne lidar, such as LVIS
(Laser Vegetation Imaging Sensor), and spaceborne
lidar, such as GLAS (Geoscience Laser Altimeter
System) on ICESat, to provide large-area maps and
inventories of vegetation structure and carbon
stocks. First-generation algorithms for
processing Echidna data focus on retrieving the
location, size, and spacing of tree trunks and on
the foliage profile of the stand. To the right
are some results for lidar scans using the
Echidna Validation Instrument (EVI), an
engineering prototype. The results include
comparison of stand height, LAI, biomass, DBH,
and stem density derived from the EVI with
contempo-raneous field measurements for sites at
Harvard Forest (MA), Howland Experimental Forest
(ME), and Bartlett Experimental Forest (NH).
Point Cloud Volume The peak intensity and its
location within each waveform were used to fill a
three-dimen-sional, gridded volume in rectangular
coordinate space to form a point cloud dataset
for each of the five lidar scans in a stand.
Considering that some leaves or branches are
always hidden by others in a single scan, and
that local terrain effects can also obscure parts
of the forest within a scan, we registered the
overlapping portion of the point clouds from the
five scans. The combined point cloud data set
then reconstructs the forest as a
three-dimensional point cloud within a central
volume.
Echidna ground-based lidar. Lidar pulses strike
a rotating mirror at an angle of 45, provide a
scan through zenith angles of 130 in a vertical
circle. As the instrument rotates on its vertical
axis, data from all azimuths are acquired.
Comparison of structural parameters retrieved by
EVI with field measurements, New England forests.
Data are averages for 5 scans or plots at each of
six sites. Standard deviations are based on
values for these 5 scans. All correlation
coefficients are significant at the 1-percent
level. 1Exception Howland Tower, 5 scans, 3
plots. 2Destructive sampling (Catovsky and
Bazzaz, 2000). 3Cohen et al., 2006. 4J. T. Lee,
University of Maine. 5A. Richardson, University
of New Hampshire.
Diameter at breast height and biomass are
retrieved well by the EVI. Different colors show
data in specific distance ranges.(Left) Because
the EVI acquires data at constant angular
resolution, EVI-retrieved diameter becomes less
accurate with distance. Although the R2 value is
only 0.620 overall, the regression slopes are so
very close to 1. Moreover, the intercept has only
a small, but very consistent value of about 0.02
m. (Right) Comparison for biomass estimation
based on EVI retrieval and field measurements.
The field biomass values are determined for each
plot using allometric equations by species
applied to each measured stem, while EVI biomass
values are calculated by a pooled allometric
equation for the two most dominant species in a
plot . At the plot level, the R2 value is 0.854,
with slopes and intercepts not significantly
different from 1 and 0, respectively.
Reconstruction of a stand based on a single EVI
scan of a giant sequoia stand, Sequoia National
Forest, near Fresno, California. Shown is a point
cloud with materials indicated by color green,
foliage brown, trunks and branches magenta,
ground. Points are labeled based on the intensity
and width of the full return waveform as
digitized by the instrument. The large trees at
either side of the image are giant sequoias.
Example of EVI data for a deciduous forest stand
at the Bartlett Experimental Forest (NH). The
upper left image is image of the mean lidar
return shown in a hemispherical projection. It
resembles a hemispherical photo. The upper left
figure shows the intensity and shape of the lidar
return which distinguishes trunks from foliage
and determines their distance from the
instrument. The bottom image is a plate carrée
projection that displays the data by azimuth
angle (x-axis) and zenith angle (y-axis).
Hardwood
Conifer
Sample design. A 1-ha area was sampled at each of
six sites (stands). The area was divided into 4
plots of 0.25 ha. We acquired EVI scans and
ground stem measurements at the center of each
plot and the center of the site (left). In each
plot, we also acquired 13 hemispherical photos
and LA-2000 measurements in the pattern shown to
the right.
A proper Auzzie echidna.
An example of surface reconstruction using three
merged lidar scans. (Left) Triangulated surface
from point cloud return data. (Right) Merged
scene of reconstructed surfaces. Colors identify
the scan from which the original point was
obtained (blue, green red).
Acknowledgements This work is supported by NASA
grant NNG06GI92G. We thank the assistance of John
Lee at Howland Experimental Forest, Audrey Barker
Plotkin at Harvard Forest, Andrew Richardson at
Bartlett Experimental Forest in the field work,
the assistance of Zhuosen Wang, Miguel Roman,
Mitchell Schull, Qingling Zhang and Shihyan Lee
in collecting field and EVI data.
Near-horizontal lidar returns can identify tree
trunks and place them accurately with respect to
the instrument. (Left) A stem map shows trunks
identified by the EVI in green. Also shown in
black are trees within a 20-m radius that were
manually located and measured. Scan points were
shifted somewhat from design locations because of
a road passing through the site.(Harvard Forest
hemlock plot.) (Right) The figure compares
plot-level retrievals of stem count density using
the EVI with measured value. The plot-level
retrievals are very accurate, with an R2 value of
0.902. Part of the high correlation is due to
three plots at the Howland Tower site, which had
2-3 times as many stems as the other plots and
anchored the high end of the regression line.
However, inspection of the remaining points still
shows a strong linear relationship.
Using multiple scans and a simple tomographic
approach, we can reconstruct a stand in 3-D. The
reconstruction opens the door to more direct
measurement of woody biomass, foliage area
volume, and structural measurements of tree
diameter and height. (See video display.)
Foliage profiles (leaf area with height)
retrieved from EVI scans at 6 sites, 5 scans per
site. Among the hardwood sites, Harvard Hardwood
is tallest, with an open understory. The Bartlett
stands are shorter with more understory. The
conifer sites show sharply peaked profiles with
little or no understory layer. The curves fit a
model of leaf area that increases with conifer
crown width until the crowns touch and intersect,
when leaf area is at its peak.
Contact Crystal Schaaf, schaaf_at_bu.edu, Alan
Strahler, alan_at_bu.edu.