Spectral and Structural Differences Between Coniferous and Broadleaf Forest derived from LIDAR and A

1
Spectral and Structural Differences Between
Coniferous and Broadleaf Forest derived from
LIDAR and AVIRIS Dar A. Roberts1, Keely L. Roth2,
Eliza Bradley3, Geoffrey G. Parker4, Philip E.
Dennison5, and Bothaina Natour6 1 Dept of
Geography, Univ. Calif. Santa Barbara, 2.
Smithsonian Environmental Research Center,
Edgewater, MD 21037-0028, 3Dept of Geography,
Univ. Utah, Salt Lake City, UH, 94119
1dar_at_geog.ucsb.edu,1 klroth_at_geog.ucsb.edu, 1
ebradley_at_geog.ucsb.edu, 2 parker_at_si.edu,
3dennison_at_geog.utah.edu,3 3bothaina.natour_at_geog.ut
ah.edu
Methods
Abstract
Study Areas
AVIRIS Processing 1) AVIRIS data were processed
to surface reflectance using radiance modeled by
MODTRAN allowing for variable amounts of column
water vapor and liquid water 2) AVIRIS
reflectance data were modeled using Multiple
Endmember Spectral Mixture Analysis (MESMA
Roberts et al., 1998) to calculate fractions of
GV, NPV and shade 3) MODTRAN modeled irradiance
were convolved against surface reflectance then
integrated across wavelengths to estimate surface
albedo (Roberts et al., 2004) 4) AVIRIS fractions
and albedo were compared to LIDAR rugosity over a
range of window sizes of 8, 12, 16 and 20 m.
Combined LIDAR and hyperspectral measures can
improve our ability to estimate carbon stocks and
fluxes, through improved maps of forest
structure, tree species and biophysical
attributes. In past research, we have
demonstrated strong relationships between small
footprint (discrete) LIDAR and canopy albedo
across an age chronosequence in coniferous
forest. We have also shown strong relationships
between LIDAR-derived height measures and biomass
in several coniferous ecosystems. In this poster
we extend this analysis to compare LIDAR and
AVIRIS-derived measures in a broadleaf deciduous
forest (SERC) to those derived from a coniferous
forest at Wind River. Analysis at SERC focused on
three main sensors, discrete LIDAR, waveform
LIDAR (LVIS) and AVIRIS. Discrete LIDAR data were
processed to generate a digital terrain model
(DTM), then used to create a digital canopy model
(DCM). A comparison between canopy height
measures derived from LVIS, and height derived
from the discrete LIDAR showed both systems to be
highly correlated. Height measures from LVIS and
discrete LIDAR were validated using canopy height
derived from a stem map and by comparison to
simulated canopy hemi-ellipsoids generated
through a model. Comparison between modeled
canopy height and LIDAR derived height were
generally good, improving when species-level
differences were incorporated into simulated
canopy hemi-ellipsoids. Correlation between
AVIRIS-derived albedo, and canopy shade at SERC
was very high. However, correlation between the
standard deviation of LIDAR-height (rugosity) and
albedo was low, contrary to previous findings at
Wind River. Likely mechanisms accounting for
differences between SERC and Wind River include
1) a lack of a large range of age classes at
SERC 2) architectural differences between
conifer and broadleaf trees and 3) differences
in the spatial scale of analysis, where
correlations are high when the data are
aggregated to stand scales, but lower when
aggregated over fixed window sizes.
Two study areas are included in this poster (Fig.
1), including western hemlock/Douglas-fir (Wind
River) and mixed broadleaf deciduous forest
(SERC). AVIRIS data include coarse resolution at
Wind River acquired in 1998 and high resolution
acquired in 2006 at SERC. LIDAR include discrete
return LIDAR at Wind River (Aeroscan) and two
LIDAR data sets at SERC, a discrete return system
acquired in 2004 and LVIS acquired in 2003. Field
data included tree height for individual trees (
200 Wind River) and a 700x700 m stem map at
SERC consisting of 9563 trees with stems gt 20 cm
DBH.
LIDAR Processing 1) Discrete LIDAR were processed
first to determine a Digital Terrain Model (DTM)
using a 16 m fixed window to locate height minima
followed by linear interpolation between minima.
A digital canopy model (DCM) was developed for
each site as the difference between LIDAR first
return (DSM) and the DTM (Fig. 2) 2) Discrete
LIDAR were reprocessed to synthesize LVIS
waveforms over footprints that matched LVIS.
Discrete LIDAR were also processed to calculate
rugosity at 8, 12, 16 and 20 m 3) Height metrics
from LVIS and discrete LIDAR were compared at
SERC 4) Height metrics from both sensors were
compared to estimated tree height from allometry
and modeled crowns using Purves et al., 2007
a)
b)
Figure 1) (a) Wind River AVIRIS image showing
the general dimensions of the LIDAR data set
(orange) (b) SERC, Maryland. Boxes mark the
locations of the three main data sets at SERC
including LVIS (white), the stem map (magenta)
and discrete LIDAR (yellow)
Figure 2) Generation of DCM from first and last
return LIDAR at SERC.
Summary
Results
1. Discrete LIDAR and LVIS
3. AVIRIS Albedo
LVIS, discrete LIDAR and AVIRIS were analyzed at
SERC Maryland, a mixed broadleaf forest and
compared to similar analysis in a coniferous
ecosystem at Wind River. Important findings
included Comparisons between height measures
derived from LVIS and discrete LIDAR at SERC
demonstrated strong correlations. The highest
correlation was observed for maximum canopy
height over a 50 m window size. Comparison
between field measured canopy height and canopy
height derived from discrete LIDAR demonstrated a
weak relationship for maximum height, but a
strong correlation for average height.
Regressions between LIDAR-derived average height
and field-derived heights improved when
species-specific models were used to synthesize
crowns from the stem map. Similar to prior
work, AVIRIS-derived albedo was highly correlated
with the GV and shade fractions at SERC.
Contrary to findings at Wind River,
AVIRIS-derived albedo was poorly correlated with
LIDAR-derived rugosity at all four spatial scales
tested.

• LVIS and discrete LIDAR produced similar
  waveforms in most cases showing a range in
  correspondence from a near perfect match in many
  instances to significant differences (Fig. 5).

• Tree height measures from discrete LIDAR and
  LVIS (RH100) proved to be highly correlated (Fig.
  3). Two window sizes were used for comparison 30
  m and 50m (Fig. 4). Highest correlations were
  observed between LVIS and discrete LIDAR for
  maximum tree height at the 50 m scale (Fig. 4a).
  Mean tree height was poorly correlated at both
  scales (Fig. 4b).

As shown previously by Ogunjemiyo et al.,
(2005) albedo derived from AVIRIS was negatively
correlated with the shade fraction and positively
correlated with the GV fraction (Fig. 8).
Brightness differences within these stands are a
product of spectral differences in crowns, crown
architecture and forest gaps.
Contrary to prior work at Wind River, albedo
was poorly correlated to LIDAR-derived rugosity
at all spatial scales of 8, 12, 16 and 20 m (Fig.
10). Differences between Wind River and SERC are
hypothesized to be due to several factors
including a) The spatial scale of analysis.
Albedo-rugosity relationships at Wind River were
calculated for discrete stands defined by the
USFS, thus suppressing within stand variability.
Albedo-rugosity at SERC was calculated from stem
maps. b) Wind river includes a wide range of
stand ages ranging from recent clear cuts to over
400 years old. SERC consists primarily of young
and intermediate aged stands. c) Wind river
analysis focused exclusively on conifers with
well defined, conical crowns. SERC analysis
includes a diversity of crown forms including
numerous overlapping crowns a majority of which
are from broadleaf deciduous trees.
b)
a)
a)
b)
b)
Figure 5) Comparison of LVIS waveforms and
synthetic waveforms generated from discrete
LIDAR. Some waveforms are nearly identical, while
others differ markedly.
Figure 8) Showing plots of shade and GV (x)
plotted against albedo (y).
4. AVIRIS Albedo and Rugosity
2. Comparison to Plot data
a)
Figure 3) DCM from discrete LIDAR and canopy
height from LVIS
Figure 4) Regression between LVIS and discrete
LIDAR at 30 m (a) and 50 m (b) for maximum height
(left) and average height (right)
LIDAR height metrics, including maximum and
mean height derived from discrete LIDAR at 50 and
100 m were compared to height measures derived
from the SERC stem map. Overall, mean height
showed a higher correlation to field data than
maximum height with highest correlations for a
100 m window size (Fig. 6). Comparison
between canopy height derived from allometry and
canopies synthesized using the ideal crown
distribution of Purves et al. (2007) demonstrated
improved correlation when species specific crown
morphology is taken into account (Fig. 7).
In prior research at Wind River, we
demonstrated a strong relationship between
LIDAR-derived roughness, calculated as the
standard deviation of LIDAR height within a fixed
window size (rugosity) and albedo and spectral
fractions (Fig. 9). As stands aged, rugosity
tended to increase, the GVF fraction and albedo
decreased and shade fraction increased.
References
Ogunjemiyo, S., Parker, G., and Roberts, D.,
2005, Reflections in bumpy terrain implications
of canopy surface variations for the radiation
balance of vegetation, IEEE Geoscience and Remote
Sensing Letters, 2(1), 90-93. Purves, DW,
Lichstein, JW Pacala, SW. (2007) Crown
plasticity and competition for canopy space a
spatially implicit model parameterized for 250
North American tree species. PLoS-ONE 2(9) e870.
doi10.1371/journal.pone.0000870 Roberts, D.A.,
Gardner, M., Church, R., Ustin, S., Scheer,
G.,and Green, R.O., 1998, Mapping Chaparral in
the Santa Monica Mountains using Multiple
Endmember Spectral Mixture Models, Rem. Sens.
Environ. 65 267-279. Roberts, D.A., Ustin, S.L.,
Ogunjemiyo, S., Greenberg, J., Dobrowski,S.Z.,
Chen, J. and Hinckley, T.M., 2004, Spectral and
structural measures of Northwest forest
vegetation at leaf to landscape scales,
Ecosystems, 7545-562
Figure 9) Showing the relationship between the
GV fraction, shade fraction, and albedo as a
function of stand age (From Ogunjemiyo et al.,
2005).
b)
Acknowledgements This research was funded in
part by Multisite Integration of LIDAR and
Hyperspectral Data for Improved Estimation of
Carbon Stocks and Exchanges, NASA Carbon Cycle
Science grant NNG05GE56G
Figure 10) Showing the relationship between the
albedo and rugosity at Wind River (a) and SERC (b)
Figure 7) Correlation between maximum and mean
tree height calculated from discrete LIDAR at 100
m compared to standard allometry (a) and
species-specific growth forms from Purves et al.
(2007) (b)
Figure 6) Correlation between maximum and mean
tree height calculated from discrete LIDAR at 50
(a) and 100 m (b) and field allometry at the same
spatial scales.