PE&RS May 2016 - page 354

The temperate forest site near
the Tumbarumba research station
(−35.39°, 148.09°) within the Bago
State Forest, New South Wales,
constitutes a relatively open Euca-
lyptus forest across complex ter-
rain; mean canopy height is 40 m
(Leuning et al., 2005; van Gorsel et
al., 2009). ALS data were collected
during November 2009 (Hop-
kinson et al., 2013) with a mean
point density of 5 pm-2 (points per
square meter). The site at Rob-
son Creek (−17.12°, 145.63°) is a
tropical rainforest on the slopes of
the Lamb range in Danbulla Na-
tional Park, within the Wet Tropics
World Heritage Area. The site is a
simple notophyll vine forest with a
closed canopy, exhibiting high spe-
cies diversity with mean canopy
heights ranging from 26 m to 40
m (Liddell, 2013). Airborne data
were collected during September
2012, acquired with a mean point
density of 5 to 6 pm-2. The site at
Watts creek (−37.69°, 145.68°) is an
open forest with a Eucalypt overstorey, dominated by Moun-
tain Ash (Eucalyptus regnans ), exhibiting canopy heights >40
m (Auscover, 2014). Airborne data were collected during April
2012, yielding a mean point density of 5 to 6 pm-2.
Airborne laser scanning (ALS) data employed in this study
were collected by Airborne Research Australia (ARA) with
a Riegl LMS-Q560 system. Raw and some derivative data
products are available through the Terrestrial Ecosystem
Research Network ( TERN)(http://www.tern.org.au). The ALS
data employed here are part of a larger TERN data collection of
13 “supersites”; however, of the data available for use in this
study only three sites exhibit collocation with GLAS data.
Methods
Canopy Heights
Two methods were used to compute canopy height from
GLAS data, based on GLA14 land data product variables (see
Table 1): RHROS (Rosette et al., 2008) and RH100 (Ni-Meister
et al., 2010; Yang et al., 2011). RHROS is defined as a constant
multiplied by the difference between the signal start eleva-
tion (i_SigBegOff), and the elevation of the center of the larger
(amplitude) of the two least elevated Gaussians (GA1,2, derived
from i_gpCntRngOff) fitted to the waveform (Equation 1). Ad-
ditionally, RH100 was employed as an alternate canopy height
measure, estimated according to Equation 2. RH100 is defined
as the difference between the signal start and the center of
least elevated of the waveform fitted Gaussians (G1).
RHROS = 1.06(i_SigBegOff – GA1,2) (1)
RH100 = i_SigBegOff – G1 (2)
While additional methods are available for estimating
canopy height from GLAS waveforms, utilizing direct and
statistical approaches (Lefsky et al., 2005; Lefsky et al., 2007;
Chen, 2010), and ground location corrections using terrain in-
dices (Rosette et al., 2008), these are often tuned locally and/
or require supplementary information. Such additions are
time consuming and inappropriate at large scales, and hence
are not employed in this study. The methods for GLAS canopy
height retrieval are illustrated in Figure 3.
Canopy Height Comparisons
RHROS and RH100 were compared with spatially concurrent ALS
derived canopy heights for each study site. Multiple GLAS /ALS
height comparisons were executed, as outlined below, and GLAS
height comparison quality was investigated according to instru-
ment, spatial/scene, and temporal aspects of GLAS data varia-
tion. Each source of data variation is investigated by compiling
and comparing statistics of height comparisons stratified by
GLAS height method (RH100 and RH ROS), GLAS laser number, GLAS
laser transmission energy, vegetation phenological state (at time
of data capture), or location (study site), with ALS equivalents.
For the investigation of laser number, GLAS data were
stratified according to GLAS laser number 1, 2, or 3. For laser
transmission energy investigations, footprints were binned
according to high and low energies, where high energy foot-
prints were defined as those with a transmission energy >28
mJ, and low energy footprints are those ≤28 mJ; where 28 mJ
threshold represents the 50th percentile of transmitted laser
energies from all GLAS footprints across all study sites. For
phenological state, GLAS footprints collected between April
and September are defined as southern hemisphere winter
conditions (WP), where data collected between October and
March are summer conditions (SP). For each of the study sites,
differences in vegetation characteristics are somewhat lim-
ited by sample size (lack of ALS and GLAS data coincidence).
However, the quality of GLAS/ALS height comparisons can
be gauged with respect to the most dominant differences in
features between sites; these are the relatively open Eucalypt
forests of Tumbarumba and Watts Creek, contrasted against
the dense subtropical rainforest at Robson Creek.
GLAS heights were compared against the 90th, 95th, 99th, and
100th (maximum) ALS height percentiles; lower percentiles are
not investigated as they are a priori not expected to correlate
well with estimates of GLAS height. ALS height percentiles
(abbreviated to pXX from here on; where XX is the percentile
value) were derived from three data sources: the all return
point cloud, the first return only point cloud, and a rasterized
Canopy Height Model (CHM). Raster CHMs were produced from
LAStools (Hug et al., 2004; Isenburg, 2011), where a 1 m pixel
radius was selected to produce 1 m spatial resolution rasters.
Height values were calculated as the percentile height of ALS
points ≥2 m above the perceived ground, (i.e., p95 is the eleva-
tion at which 95 percent of points in that pixel are between 2
Figure 3. Visualization of the two GLAS canopy height retrieval methods employed in this study.
Methods are illustrated with respect to (a) the returned waveform, and (b) the waveform Gauss-
ian decomposition. RH100 always utilizes the least elevated Gaussian (greatest relative time)
as the ground, whereas RHROS screens the two least elevated Gaussians for that which has the
greatest amplitude to indicate ground.
354 May 2016 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING