PE&RS September 2014 - page 863

September 2014
Generating Pit-free Canopy Height
Models from Airborne Lidar
Anahita Khosravipour, Andrew K. Skidmore, Martin Isenburg, Tiejun Wang, and Yousif A. Hussin
Canopy height models (
) derived from lidar data have
been applied to extract forest inventory parameters. How-
ever, variations in modeled height cause data pits, which
form a challenging problem as they disrupt
ness, negatively affecting tree detection and subsequent
biophysical measurements. These pits appear where laser
beams penetrate deeply into a tree crown, hitting a lower
branch or the ground before producing the first return. In
this study, we develop a new algorithm that generates a
raster, by using subsets of the lidar points to
close pits. The algorithm operates robustly on high-den-
sity lidar data as well as on a thinned lidar dataset. The
evaluation involves detecting individual trees using the
and comparing the findings to those achieved
by using a Gaussian smoothed
. The results show that
our pit-free
derived from high- and low-density lidar
data significantly improve the accuracy of tree detection.
The use of airborne Light Detection and Ranging (lidar) has
been increasing in forestry. Lidar is capable of providing accu-
rate three-dimensional information on forest structure (Lim
, 2003a), contributing significantly to the improved accu-
racy of forest inventories (Magnussen
et al.
, 2010; Yu
et al.
2011) and subsequent biophysical parameters such as biomass
et al.
, 1988; Popescu, 2007).
Typically, a lidar-derived Canopy Height Model (
) or a
normalized Digital Surface Model (n
) is used for extract-
ing relevant forest inventory information, such as detecting
single trees for subsequent height estimation and crown
delineation (Bortolot and Wynne, 2005; Forzieri
et al.
, 2009).
represents absolute canopy height above ground, and
it is typically calculated by interpolating the first return lidar
points and determining their height above a digital terrain
model. (Hyyppä
et al.
, 2008; Van Leeuwen
et al.
, 2010). Tree
height measurement and crown delineation mainly rely on
the identification of local maxima, with each local maximum
corresponding to the location of an individual treetop and
the surrounding segments forming the tree crown (Véga and
Durrieu, 2011). Therefore, to be able to extract relevant struc-
tural parameters of trees (e.g., tree height) the correct location
of single trees in the
is of fundamental importance (Chen
et al.
, 2006; Persson
et al.
, 2002; Yao
et al.
, 2012). While some
researchers have tried to find local maxima directly in the li-
dar points (Li
et al.
, 2012), most operational users of lidar first
calculate a raster
from the first return lidar points and
then extract local maxima from that raster
et al.
2008; Lim
et al.
, 2003b).
The main challenges faced in treetop detection are com-
mission errors (falsely detected trees) and omission errors (un-
detected trees) (Hosoi
et al.
, 2012; Pouliot
et al.
, 2005). These
errors are mainly attributed to natural variation in tree crown
size (Pitkänen
et al.
, 2004) as well as to height irregularities
within individual tree crowns in the input
et al.
2006). To address natural variation in crown size, researchers
have developed processing methods that adapt to the crown
(object) size. Pitkänen
et al.
(2004) developed and tested three
different adaptive methods for individual tree detection based
on canopy differences. Wulder
et al.
(2000) proposed the use
of a local maxima filter with variable window sizes. However,
if the selected window size is smaller or larger than the crown
size, then the commission or omission error, respectively, will
increase. In order to select the correct window size, Popescu
and Wynne (2004) introduced an adaptively varying window
technique, based on the idea that a moving local maxima filter
should be adjustable to an appropriate width to account for
different crown sizes.
To address irregularities in crown height, a number of
researchers have suggested pre-processing
to reduce
commission and omission errors (Bortolot and Wynne, 2005;
et al.
, 2003; Chen
et al.
, 2006; Solberg
et al.
, 2006).
Irregularities in canopy surface elevation, also called “data
pits,” form a challenging problem due to their disruptive
influence on a
, reducing accuracy in tree detection and
subsequent biophysical measurements (Ben-Arie
et al.
, 2009;
Gaveau and Hill, 2003; Zhao
et al.
, 2009). For example, Sham-
et al
. (2013) indicated that data pits may significantly
affect the estimation of structural forest parameters, especial-
ly basal area and stand volume. Since the processing of raw
lidar point clouds into a meaningful raster is a composition
of many different procedures, there is no unified agreement
on the cause of data pits. Axelsson (1999) found that some in-
formation from raw point clouds with similar
and different
values is lost when the points are interpolated
into a raster. Such lost data become significant when multiple
echoes are registered in a forested area. Ben-Arie
et al.
and Véga and Durrieu (2011) stated that the problem of data
pits was due to laser scanning processing and/or postpro-
cessing of lidar point clouds. Data pits may also occur during
classification of lidar point clouds into ground and non-
ground points when creating a Digital Surface Model (
) or
a Digital Terrain Model (
), depending on the classification
Anahita Khosravipour, Andrew K. Skidmore, Tiejun Wang,
Yousif A. Hussin are with the Department of Natural Resourc-
es, Faculty of Geo-Information Science and Earth Observa-
tion, University of Twente, P.O. Box 217, 7500 AA Enschede,
The Netherlands (
Martin Isenburg is with Rapidlasso GmbH, Friedrichshafener
Straße 1, 82205 Gilching, Germany.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 9, September 2014, pp. 863–872.
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.9.863
811...,853,854,855,856,857,858,859,860,861,862 864,865,866,867,868,869,870,871,872,873,...914
Powered by FlippingBook