A Geometric Method for Wood-Leaf Separation
Using Terrestrial and Simulated Lidar Data
Shengli Tao, Qinghua Guo, Shiwu Xu, Yanjun Su, Yumei Li, and Fangfang Wu
Abstract
Terrestrial light detection and ranging (lidar) can be used to
record the three-dimensional structures of trees. Wood-leaf
separation, which aims to classify lidar points into wood
and leaf components, is an essential prerequisite for deriving
individual tree characteristics. Previous research has tended
to use intensity (including a multi-wavelength approach) and
waveform information for wood-leaf separation, but use of the
most fundamental information from a lidar point cloud, i.e.,
the x-, y-, and z- coordinates of each point, for this purpose
has been poorly explored. In this study, we introduce a geo-
metric method for wood-leaf separation using the x-, y-, and z-
coordinates of each point. The separation results indicate that
first-, second-, and third-order branches can be extracted from
the raw point cloud by this new method, suggesting that it
might provide a promising solution for wood-leaf separation.
Introduction
Light detection and ranging (lidar) is an active remote-sensing
technology that can be used to record three-dimensional (3D)
information on objects (Wehr and Lohr, 1999). According to
the platform, lidar can generally be categorized into three
types: satellite-based lidar, airborne lidar, and terrestrial lidar
(T-lidar) (van Leeuwen and Nieuwenhuis, 2010). Satellite-
based lidar and airborne lidar can capture large-scale 3
D
information on forest structure, and have advanced forestry
research in a range of applications, including biomass calcula-
tion (Popescu, 2007), tree height measurement (Means
et al
.,
2000; Li
et al
., 2012), and carbon mapping (Asner
et al
., 2010).
However, only limited information can be obtained at the tree
or branch scale using satellite-based lidar and airborne lidar.
A complementary technique, T-lidar, operates underneath the
canopy, providing a revolutionary way of quantifying individ-
ual tree characteristics, with details and accuracy that satellite-
based lidar and airborne lidar cannot match (Ducey
et al
.,
2013). During the past decade, T-lidar has attracted increasing
attention, because of its accuracy and flexibility. Significant
progress has been made with respect to species identification
(Dassot
et al
., 2011), diameter at breast height (
DBH
) and leaf
area index (
LAI
) retrieval (Henning and Radtke, 2006; Hosoi
and Omasa, 2006; Moorthy
et al
., 2008), plant biomass and
volume calculation (Lefsky and McHale, 2008; Ku
et al
., 2012),
and forest structural parameter extraction (Mass
et al
., 2008).
Some of the forest characteristics that T-lidar can acquire
can be directly estimated at the individual tree level, such
as stem density,
DBH
, and tree height. Simonse
et al
. (2003)
and Schilling
et al
. (2012) used the two-dimensional (2
D
)
Hough transform to detect trees and to measure the
DBH
after
projecting 3
D
points onto a 2
D
grid. Mass
et al
. (2008) ap-
plied morphological techniques on lidar points for automatic
determination of
DBH
, tree height, and 3
D
stem profiles. Liang
et al
. (2012) presented an automatic circle-fitting procedure
based on the spatial distribution properties of the laser points
for stem mapping. However, there are certain tree measure-
ments for which wood-leaf separation is an essential pre-
requisite, as shown by the following examples. Wood points
are needed to calculate the number of branches, the length
and radius of each branch, and the total wood volume of a
tree (including the trunk, branches, and twigs) (Lefsky and
McHale, 2008; Kelbe
et al
., 2013). Leaf points are needed to
calculate the
LAI
and leaf area density, using the Beer-Lambert
law or the Warren-Wilson contact frequency method (Beland
et al
., 2014). For better modeling of the 3
D
radiative transfer
properties of vegetation canopies, the wood and leaf parts of
each tree should be separated and assigned different spectral
reflectances (Widlowski
et al
., 2014). Computer visualization
of trees requires wood points and leaf points to be separated
and visualized using different techniques, because the shapes
of branches and leaves differ significantly (Xu
et al
., 2007).
Applications of these techniques in forestry would improve
our knowledge in a range of areas, from exploring the allome-
tric relationships hidden in tree branch networks to quantify-
ing the structural complexity of forests (Bentley
et al
., 2013).
Wood-leaf separation for T-lidar data is technologically
challenging. Existing approaches use intensity information,
a multi-wavelength approach, or waveform information for
wood-leaf separation. The intensity approach is based on
identifying an appropriate intensity value as a threshold to
differentiate wood and leaves (Pfennigbauer and Ullrich,
2010; Wu
et al
., 2013; Beland
et al
., 2014). This approach as-
sumes that there are significant differences among the optical
properties of different components of a tree at the operating
wavelength of the lidar system. However, the wavelengths and
powers of different lidar systems are not consistent (Lefsky
et
al
., 2002). Successful intensity-based wood-leaf segmentation
using one lidar system does not guarantee successful segmen-
tation using other lidar systems. Furthermore, trees of different
species respond differently to the lidar wavelength, and this
means that the intensity approach cannot be used for some
tree species. Recently, a multi-wavelength approach for wood-
leaf separation was reported (Li
et al.
, 2013). They merged
the data obtained using the shortwave-infrared (
SWIR
) laser
Shengli Tao is with the Peking University, Department of
Ecology, College of Urban and Environmental Sciences,
Beijing 100871, China.
Qinghua Guo and Yanjun Su are with the Sierra Nevada
Research Institute, School of Engineering, University of
California at Merced, CA 95343 (
.
Shiwu Xu is with the China University of Geosciences,
Faculty of Information Engineering, Wuhan 430074, China.
Yumei Li and Fangfang Wu are with the State Key Laboratory
of Vegetation and Environmental Change, Institute of Botany,
Chinese Academy of Sciences, Beijing 100093, China.
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 10, October 2015, pp. 767–776.
0099-1112/15/767–776
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.10.767
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2015
767
