(Broich
et al
., 2014). Hence phenological investigations were
carried out during seasonal extremes: summertime phenology
(
SP
), and wintertime phenology (
WP
)
.
The three canopy ecosystems studied demonstrate vari-
able terrain and vegetation characteristics. This enables an
additional sensitivity investigation of
GLAS
canopy heights,
as the shape of the returned waveform is influenced by such
characteristics. In particular, terrain has been noted to create
problems for
GLAS
waveform interpretation due to vegetation
and ground returns originating from similar elevations (Lef-
sky
et al.
, 2005; Chen, 2010; North
et al
., 2010; Simard
et al
.,
2011; Los
et al
., 2012; Rosette
et al
., 2013). These issues are
well documented, and are not discussed here
.
Given similar observations may be arrived at from different
means (i.e., can be equifinal), it is impossible to completely
control for each of the above noted influences (laser number,
energy state, phenology, and site). Moreover,
GLAS
/
ALS
height
relationships are potentially susceptible to variation resulting
from different causes than those listed. For example, changing
either the angle of incidence of laser pulses or the orientation of
leaves within a canopy might alter the
ALS
point cloud frequen-
cy distribution or the shape of a
GLAS
waveform. An occurrence
of this may be noted for Eucalyptus as its leaves hang vertically
(ANBG, 2014), hence when viewed from nadir, ground returns
are more probable than from an off-nadir viewing geometry,
where ground occlusion is more probable. Evaluating such
influences is outside the scope of the presented analysis but is
noted as a potential source of noise, and serves to emphasize
that no such experiments can be
perfectly
controlled
.
Based on the literature, Laser 3 is expected to perform bet-
ter than Lasers 1 or 2. It is hypothesized that constraining the
GLAS
data used in future modeling steps to summertime phe-
nological conditions and high energy laser transmissions will
supplement laser 3’s expected superior performance. Sum-
mertime conditions are expected to yield improved results as
more leaf area provides a larger cross section for incident light
to be reflected back to the detector and thus better represent
canopy structure and height. Small cross sections can result
in the intersected surface not reflecting sufficient energy to
be registered above detector noise levels. Similarly, higher
energy emissions should yield a better signal to noise ratio.
A visualization of these concepts is illustrated in Figure 1,
with hypothetical waveform returns. The analysis presented
will test these hypotheses for a range of
GLAS
and
ALS
height
metrics and present quantifications of the sensitivity within
each evaluation stratum.
As part of a larger study to model Australian canopy at-
tributes from
GLAS
data, the current study investigates sources
of variation in the
GLAS
/
ALS
canopy height inter-relationship
associated with instrument, temporal and site-level attributes
across three Australian ecosystems.
GLAS
system-induced
variations in canopy height associated with footprints
captured by different onboard lasers (number 1, 2, or 3) and
laser energy states (high or low) are investigated. Environ-
ment-based variations due to canopy phenology conditions
(summer versus winter) are also studied. Finally, variations
in
GLAS
/
ALS
height inter-relationships are investigated with re-
spect to unique environmental characteristics (such as terrain,
species, and canopy structure) at each study site. In order to
reach this end, two
GLAS
derived height methods, RH
100
(Ni-
Meister
et al.
, 2010; Yang
et al
., 2011) and RH
ROS
(Rosette
et
al
., 2008), are compared to
ALS
point cloud frequency distri-
bution analogues (described below)
.
This work aims to generate an “optimized”
GLAS
canopy
height dataset based on
ALS
height data, and prioritize which
of the evaluated stratum yields the greatest improvement in
GLAS
canopy height estimates with respect to
ALS
equivalents.
Optimized
GLAS
canopy height estimates are essential to ex-
tend
ALS
-based site-level canopy attribute models to the scale
of the Australian landmass and beyond.
Data
Waveform Data
In this study the
ICESat GLAS
land data (GLA14) product,
release 33 (Zwally
et al
., 2011), was employed over the
Figure 1. Conceptual illustration of the anticipated effect of phenological state, and laser transmission energy on returned GLAS wave-
forms. Summertime returns are less likely to miss canopy tops due to the increased cross section provided by greater leaf area, where
the converse is true for winter. High energy returns exhibit greater returned signal strength in comparison to low energy returns, notable
in the return signal magnitude. The suspected worst case is in (b) ii) where the canopy top is missed due to both low energy and lack of
cross section from little/no leaf mass
352
May 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
