PE&RS May 2016 - page 361

Dense canopy cover in combination with some topographic
relief (~10° slope) can obstruct the ground in
GLAS
signal
returns, making canopy heights irretrievable (Rosette
et al.,
2013). In addition, terrain complexity introduces uncertainty
into canopy height estimates when ground returns occur at sim-
ilar elevations to those of vegetation returns, leading to signal
mixing and ambiguous ground elevations (Neuenschwander
et al.
, 2008; Lin
et al.
, 2010; Rosette
et al.
, 2010; Mahoney
et
al.
, 2014). Results from the current study regarding unique site
analysis, concur with existing literature that suggests more
homogeneous forests with simple terrain yield best estimates of
canopy height (Lefsky
et al.
, 2002; North
et al.
, 2010)
.
The improvements noted between control and optimal
data demonstrate the value of the current work. The im-
portance of each stratification is also indicated, where laser
number selection is noted as the most important, and the state
of phonological conditions under which data were acquired is
least important. However, the importance of the latter is likely
to be diminished here as Australia’s vegetation rarely com-
pletely defoliates, unless exposed to extreme conditions
.
While an attempt has been made to control the distribution
of data within a given stratification across the other two strati-
fication tests (as noted in Table 2), inevitably some instances
of data mixing occur. This is unavoidable, particularly as co-
incidence between
GLAS
and
ALS
data is limited. Reducing the
sample data population further in pursuit of equally distribut-
ed data across all tests, would add further integrity to results,
but at the cost of useable data. Given the current, relatively
well balanced distribution of data across each stratification,
the findings of the study are not expected to change even if a
perfect data distribution could be achieved.
Conclusions
ICESat
/
GLAS
represents the only multi-year, near global satel-
lite lidar dataset collected to date, affording opportunities to
measure and monitor vegetation within the global biosphere.
GLAS
data can be employed as primary observations to model
further vegetation metrics. Such measurements can then be
employed as inputs in, for example, climate simulations.
Hence reliable canopy height estimates are required to mini-
mize error propagation throughout subsequent modeling steps
.
This study has filtered and investigated
GLAS
measure-
ments according to the instrument, temporal sampling, and
site-level influences in order to demonstrate that derived
canopy height data can be optimized to retrieve more accu-
rate results when controlled with
ALS
data.
GLAS
data were
filtered according to laser number, phenological state, and
laser transmission energy; in addition, data were investigated
as a function of study site to demonstrate
GLAS
data quality
vary with site canopy structure and terrain characteristics.
This allowed the ecosystem over which
GLAS
performs best
(within this study) to be noted; in this instance the Eucalypt
forest at Watts Creek. Such a result holds significance within
Australia, as approximately 75 percent of the country’s forests
are classified as Eucalypt (DAFF 2014)
.
Investigated stratifications indicate that laser number selec-
tion appears to yield the most improvement to
GLAS
estimates
of canopy height with respect to
ALS
equivalents, whereas
stratifying by the phenological state under which
GLAS
data
were acquired has the least effect. Best comparisons are ob-
tained from
GLAS
data that are
high emission energy
(>28 mJ)
measurements from
laser 3
, acquired during
summertime phe-
nological
conditions. These criteria produce an optimal data-
set given the land cover conditions encountered across Aus-
tralia; however, the effect of particular land cover types, such
Figure 9. Comparison of canopy heights derived from GLAS (RH
ROS
) and ALS (all return p95) data for (a) a control (non-optimized)
dataset, and (b) an optimized dataset (which resulted from the current study).
T
able
8. S
ummary
S
tatistics
for
C
ontrol
and
O
ptimized
GLAS RH
ROS
D
ata
when
C
ompared
to
A
ll
R
eturn
p
95 D
ata
from
the
ALS.
RH
ROS
– All Return p95
Data
N RMSE R
2
F
20
F
B
Control
480
9.55
0.51 0.64 0.06
Optimized 110
8.10
0.69 0.68 0.07
T
able
9. S
ummary
of
the
P
ercentage
C
hange
in
C
omparison
S
tatistics
from
the
B
est
C
ontrol
D
ataset
,
and
the
B
est
R
esulting
D
atasets
from
E
ach
T
est
C
onducted
W
ithin
E
ach
S
tratification
. N
ote
: P
ositive
P
ercentages
I
ndicate
an
increase
in
value
,
whereas
N
egative
P
ercentages
I
ndicate
a
decrease
in
value
.
Dataset
Δ
RMSE [%]
Δ
R
2
[%]
Δ
F
20
[%]
Δ
|F
B
|[%]
Laser Number
-62
+55
+7
-50
Phenology
-7
+16
-18
+100
Transmission
Energy
-54
+80
-48
-50
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
May 2016
361
299...,351,352,353,354,355,356,357,358,359,360 362,363,364,365,366,367,368,369,370,371,...390
Powered by FlippingBook