Top-Down Analysis of Forest Structure and Biogeochemistry Across Hawaiian Landscapes 1

Abstract

Technical and analytical improvements in aircraft-based remote sensing allow synoptic measurements of structural and chemical properties of vegetation across whole landscapes. We used the Carnegie Airborne Observatory, which includes waveform light detection and ranging (LiDAR) and high-fidelity imaging spectroscopy, to evaluate the landscapes surrounding four well-studied sites on a substrate age gradient across the Hawaiian Islands. The airborne measurements yielded variations in ground topography, canopy height, and canopy nitrogen (N) concentration more accurately than they could have been obtained by any reasonable intensity of ground-based sampling. We detected spatial variation in ecosystem properties associated with the properties of different species, including differences in canopy N concentrations associated with the native species Metrosideros polymorpha and Acacia koa, and differences brought about by invasions of the biological N fixer Morella faya. Structural and chemical differences associated with exotic tree plantations and with dominance of forest patches by the native mat-forming fern Dicranopteris linearis also could be analyzed straightforwardly. This approach provides a powerful tool for ecologists seeking to expand from plot-based measurements to landscape-level analyses.

Full text

359
Top-Down Analysis of Forest Structure and Biogeochemistry
across Hawaiian Landscapes
1
Peter M. Vitousek,
2,6
Michael A. Tweiten,
3
James Kellner,
4
Sara C. Hotchkiss,
3
Oliver A. Chadwick,
5
and Gregory P. Asner
4
Abstract: Technical and analytical improvements in aircraft-based remote sens-
ing allow synoptic measurements of structural and chemical properties of vege-
tation across whole landscapes. We used the Carnegie Airborne Observato ry,
which includes waveform light detection and ranging (LiDAR) and high-fidelity
imaging spectroscopy, to evaluate the landscapes surrounding four well-studied
sites on a substrate age gradient across the Hawaiian Islands. The airborne
measurements yielded variations in ground topography, canopy height, and can-
opy nitrogen (N) concentration more accurately than they could have been
obtained by any reasonable intensity of ground-based sampling. We detected
spatial variation in ecosyst em properties associated with the properties of dif-
ferent species, including differences in canopy N concentrations associ ated
with the native species Metrosideros polymorpha and Acacia koa, and differences
brought about by invasions of the biological N fixer Morella faya. Structural
and chemical differences associated with exotic tree plantations and with domi-
nance of forest patches by the native mat-forming fern Dicranopteris linearis also
could be analyzed straightforwardly. This approach provides a powerful tool for
ecologists seeki ng to expand from plot-based measurements to landscape-level
analyses.
Substantial effort has gone into deter-
mining and analyzing ecosystem properties
and processes on a long substrate age gradient
(LSAG) across the Hawaiian Archipelago
(Crews et al. 1995, Kitayama and Mueller-
Dombois 1995, Riley and Vitousek 1995,
Chadwick et al. 1999, Herbert and Fownes
1999, Vitousek 2004), as a model for under-
standing long-term soil and ecosystem devel-
opment. As with other gradient-based studies,
the LSAG research has focused on particu-
lar sites that were selected to represent each
substrate age. This focus on particular sites is
useful because sources of variation other than
substrate age (topography, human distur-
bance, biological invasion) can be controlled
through careful site selection; moreover,
where long-term experiments (Vitousek and
Farrington 1997, Harrington et al. 2001) or
expensive or repeated measurements (Hobbie
and Vitousek 2000, Wiegand et al. 2005) are
carried out, it may also be the only pra ctical
approach. However, this site-based approach
does not account for spatial variation in eco-
system properties on landscapes, whether
that variation arises from systematic chan ges
in landscapes that occur as a consequence of
substrate age or from other processes.
Remote sensing offers a means for evaluat-
ing landscapes synoptically, rather than site
Pacific Science (2010), vol. 64, no. 3:359–366
doi: 10.2984/64.3.359
: 2010 by University of Hawai‘i Press
All rights reserved
1
Research supported by NSF grants DEB-0716852,
-0715593, -0717382, and -0715674; NASA Terrestrial
Ecology Program-Biodiversity grant NNG-06-GI-87G;
and the Carnegie Institution. The Carnegie Airborne
Observatory is made possible by the W. M. Keck Foun-
dation and William Hearst III. Manuscript accepted 1
September 2009.
2
Department of Biology, Stanford University, Stan-
ford, California 94305 (e-mail: vitousek@stanford.edu).
3
Department of Botany, Birge Hall, University of
Wisconsin, Madison, Wisconsin 53706.
4
Department of Global Ecology, Carnegie Institu-
tion of Washington, Stanford, California 94305.
5
Department of Geography, University of Califor-
nia, Santa Barbara, California 93106.
6
Corresponding author.

by site. However, until recently most remote
sensing methods have been insensitive to any
but extreme variations in the structure and
biogeochemistry of ecosystems. Recent ad-
vances in light detection and ranging
(LiDAR) (Lefksy et al. 2002) and in high
spectral and spatial resolution imaging spe c-
troscopy (High-Fidelity Imaging Spectros-
copy, or HiFIS) (Smith et al. 2002, Ustin
et al. 2004) have made it possible to measure
certain structural and chemical properties
of ecosystems across landscapes. In previous
studies, we have used LiDAR and HiFIS
separately to evaluate the landscape-level in-
fluence of biological invasions on the three-
dimensional structure and canopy nitrogen
(N) concentrations of Hawaiian forests (As-
ner and Vitousek 2005, Asner et al. 2008),
and of erosion/topography on phosphorus
biogeochemistry in the oldest landscape on
the gradient (Porder et al. 2005a).
An earlier study (Vitousek et al. 2009) used
LiDAR and HiFIS in combination to evalu-
ate aspects of forest structure and biogeo-
chemistry on the landscape level in four
sites across the Hawaiian substrate age gradi-
ent. That study demonstrated that although
the same patterns in canopy height (taller
in the two younger sites) and canopy N con-
centrations (higher in the two intermediate-
aged sites) occur on the landscape scale as
were observed in ground-based rese arch on
the focal sites in each landscape, the long-
term research sites were biased toward areas
with taller-than-average canopies within each
landscape, particularly on the two older
substrates. That study also demonstrated
that stratification of landscapes into cover
classes defined by the influence of invasion,
erosion/topography, human disturbance, and
other processes accounts for a substantial
proportion of the variation in canopy height
and N concentration within some of the
landscapes.
In this paper, we use the same LiDAR and
HiFIS information together with ground-
based sampling to understand aspects of for-
est structure and biogeochemistry in 2 km by
2 km regions around each of the four age-
gradient sites. The earlier paper focused on
landscape-level changes in ecosystem proper-
ties across the substrate age gradient, but here
we concentrate on using the imagery to un-
derstand source s of variation in ecosystem
structure and composition within each re-
gion. We thereby seek to place each of the
long-term research sites into the context of
the landscape that encompasses it.
materials and methods
Study Sites and Landscapes
The 4-million-year LSAG across the Hawai-
ian Archipelago consists of a series of sites
that are matched in elevation (1,200 m), an-
nual precipitation (2,500 mm), topography
(constructional surface of shield volcanoes),
vegetation (>80% basal area is the native
tree Metrosideros polymorpha Gaud), and to
the extent possible distur bance history (none
has been cleared by people). The gradien t is
described in detail in Vitousek (2004), and
the peculiarities of particular sites are de-
scribed in supplemental material accessible
at http://www.stanford.edu/group/Vitousek/
princetonbook.html. Here we focus on 2 k m
by 2 km landscapes surrounding four of the
focal sites on the LSAG: the 0.3-thousand-
yr-old (kyr) Thurston, 20 kyr Laupa
¯
hoehoe,
150 kyr Kohala, and 4,100 kyr Ko
¯
ke‘e sites.
The first three sites are on the island of
Hawai‘i; the last is on the island of Kaua‘i.
Remote Sensing
Methods for the remote sensing analyses are
described in Vitousek et al. (2009). Briefly,
we used the Carnegie Airborne Observatory
(CAO), a system designed to determine
chemical and structural properties of vegeta-
tion from the air (http://cao.stanford.edu)
(Asner et al. 2007). The CAO combines three
instrument subsystems into a single airbo rne
package: (1) High-Fidelity Imaging Spec-
trometer (HiFIS), here the Airborne Visible
and Infrared Imaging Spectrometer (AVIRIS)
(Green et al. 1998); (2) Waveform Light De-
tection and Rangi ng (LiDAR) scanner; and
(3) Global Positioning System-Inertial Mea-
surement Unit (GPS-IMU). We use d meth-
ods described by Asner et al. (2007) to match
360 PACIFIC SCIENCE
.
July 2010

HiFIS and LiDAR data in three-dimensional
(3-D) space. The system was operated from
January to February 2007 at an altitude aver-
aging 3.0 km above ground level, thus pro-
viding spectroscopic measurements at 3.0 m
spatial resolution and LiDAR spot spacing
(postings) of 1.0–1.5 m, depending upon the
site. The GPS-IMU data were combined
with the laser range data to determine the
3-D location of the laser returns. From the
LiDAR point cloud data, a physically based
model was used to estimate top-of-canopy
and ground digital elevation models (DEM)
using REALM (Optech Inc., Toronto, Can-
ada) and Terrascan/Terramatch (Terrasolid
Ltd., Jyva
¨
skyla
¨
, Finland) software packages.
Vegetation height was then calculated as the
difference between the top-of-canopy and
ground DEM.
The HiFIS data were converted to at-
sensor radiances by applying radiometric cor-
rections developed during sensor calibration
in the laboratory. Apparent surface reflec-
tance was then derived from the radiance
data using an automated atmospheric correc-
tion model, ACORN 5LiBatch (Imspec LLC,
Palmdale, California), with settings described
by Asner et al. (2008 ). Upper-canopy leaf N
concentration was estimated from the HiFIS
data using the visible (450–690 nm) and the
shortwave-infrared (1,500–2,400 nm) wave-
length regions, as described by Asner and
Vitousek (2005).
Field Observations
The processed images resulting from LiDAR
and HiFIS acquisitions provided a synoptic
view of the topography, canopy height, and
upper-canopy N concentration across 2 km
by 2 km landscapes centered on each of the
focal sites. Formal validation of the accuracy
of these products was carried out separately
(see Asner and Vitousek [2005] for canopy
N, Asner et al. [2008, 2009] for vegeta tion
height); here, we sought to use ground-based
observations to understand sources of the
spatial variation detected by remote sensing.
Our field sampling was primarily descrip-
tive; we printed the topographic, tree height,
and canopy N images from each area with
an overlay of GPS coordinates and selected
sample points in advance. For these sample
points, we sought representation of both fre-
quent and unusual combinations of remotely
detected topography, canopy height, and can-
opy N within each landscape. We took the
images into the field and used GPS to locate
the preselected points. Additional sample
points were selected in the field when condi-
tions not encompassed by the preselected
points wer e encountered. In most cases, we
knew what to expect from the imagery; for
example, very tall homogeneous patches of
forest on a consistent slope likely represent
plantations of exotic trees, and our field ob-
servations confirmed those expectations. In
other cases we did not know what to expect
when we reached the selected points, and the
ground observations were essential to under-
standing the remote observations.
At each point, we recorded canopy charac-
teristics (dominant canopy and understory
species, visual estimates of structure), ob-
tained GPS coordinates, and obtained and
archived two digital photographs, one taken
vertically that showed the canopy and one
taken horizontally showing the understory.
This approach allowed us to walk through
the sites both virtually (observing the images)
and literally (with the images in hand); it pro-
vides a powerful way to identify and evaluate
landscape-level variation in forest ecosystems.
Both the remote images and the ground-
based information and photographs are ac-
cessible at http://www.stanford.edu/group/
Vitousek/lsaglandscape.htm. We encourage
readers to scroll through the images and pho-
tographs and develop their own sense for
these landscapes.
Patterns in Topography and Forest Canopi es
The LiDAR-derived topographies of the 2
km by 2 km landscapes across the substrate
age gradient a re illustrated in Plates I–IV;
the same informat ion is presented in Vitousek
et al. (2009), but the figures are organized and
used differently there, and additional informa-
tion is provided. The topographic images pro-
vide remarkable resolution, even under forest
canopies, and in some cases they enabled us
Aircraft-Based Analysis of Forest Landscapes
.
Vitousek et al. 361

to identify and explain topographic features
that we had not interpreted correctly in 20–
25 yr of fieldwork in and around these sites.
Topography is an important source of
variation in forest ecosystems; fluvial erosion
and deposition influence nutrient availability
and can cause differences in foliar chemistry
among sites (Porder et al. 2005a,b), and to-
pography can shelter stands from wind dis-
turbance, thereby affecting their structure
and chemistry. Also, steep topography can in-
terfere with our analyses of both tree height
and canopy N: the former because the uphill
and downhill sides of a given tree yield dif-
ferent returns to the LiDAR (Su and Bork
2006); the latter because steep slopes create
shadows that make it difficult to derive bio-
chemically meaningful spectra (Roberts et al.
1993, Kupiec and Curran 1995). Accordingly,
we classified steep slopes (>20 degrees) as
erosionally or volcanically derived but then
masked them out of further analyses.
We then evaluated canopy heights from
LiDAR and N concentrations from HiFIS
(Plates I–IV ) across the landscapes. In com-
bination with topography, and with our
ground-based observations of a number of
points in each landscape, we defined areas
within each 2 km by 2 km landscape that rep-
resented little-invaded native forest growing
on constructional geomorphic surfaces, and
areas that were influenced substantially by bi-
ological invasions (‘‘substantially’’ defined by
areas in which invaders influenced remotely
detected canopy properties); fluvial erosion
or deposition and steep or recent volcanic to-
pography (from the image of topography);
human disturbance or land use (based on
remotely detected canopy prop erties supple-
mented by ground observations); steep cli-
matic gradients within a landscape; or other
sources of variation (Plates I–IV ).
results and discussion
Thurston Landscape (0.3 kyr)
The topography surrounding the youngest
region on the LSAG is constructional; all of
the relief visible in the topographic image
(Plate I) is volcanic in origin. Across most of
this landscape, the substrate is pa
¯
hoehoe lava
overlain with cinder. The major topographic
features are the wall and floor of Kı
¯
lauea Iki
Crater to the west of the focal site (points 14
and 15; the point numbers are shown in the
upper left section of each plate and online as
discussed earlier), and a nearby series of pits
that resulted from the collapse of portions
of the ceiling of the large underlying Thur-
ston Lava Tube. Away from the 50-yr-old
lava lake in Kı
¯
lauea Iki Crater, most of the
area is covered by forests of intermediate
height (10–20 m) (Plate I); short er canopies
occur in active and abandoned pastures in
the northern part of the image (34, 37), in
smaller areas of stand-level dieback north and
east of the focal site (6, 8), and in areas with
shallow soils that ground observations show
to be codominated by the mat-forming fern
Dicranopteris linearis Underw. in the south-
western part of the image (18).
The major source of variation in canopy N
across this landscape (Plate I) is biological in-
vasion by the actinorrhizal N-fixing tree Mor-
ella (Myrica) faya (Aiton) Wilbur (points 7,
21, 32, and others). Canopy N levels are in-
creased in invaded stands, but canopy heights
are similar and have the same level of vari-
ability as in intact stands (Plate I). A planted
stand of the native N-fixer Acacia koa (point
35) also created an area of high canopy N.
Overall, biological invasion, human land
clearing, and volcanic activity affect substan-
tial fractions of the Thurston landscape
(Plate I).
Laupa
¯
hoehoe Landscape (20 kyr)
The focal site for the second-youngest land-
scape is located on small interfluve in deep
volcanic ash soils (Plate II). The area to the
south and east of this site is covered by a
P5,000-yr-old Mauna Kea lava flow (Wolfe
and Morris 1996). Most of the ‘a‘a
¯
lava flow
surface is little dissected by erosion; its topog-
raphy largely reflects lava pressure ridges and
flow channels. However, the portion of this
flow within P300 m of the focal site is man-
tled with tephra tens of centimeters thick, as
can be seen from the subdued flow topogra-
phy there (Plate II) and as is confirmed from
362 PACIFIC SCIENCE
.
July 2010

ground observations. We wonder if in fact
there are two flows here , with an older one
closer to the focal site partially overridden by
the 5,000-yr-old flow farther to the south and
east, or alternatively if the portion of the flow
nearest the focal site represents an early por-
tion of the flow that was covered by later
tephra from the same eruptive episode. With-
in the northwesterly portion of the landscape
that is underlain by deep er ash soils, the
drainage network is fine but sha llow, with
little more than 5 m of local relief. Older
landscapes have deeper but sparser drainage
networks, as discussed in the next sections.
Tree heights within the Laupa
¯
hoehoe
landscape vary substantially, with larger trees
downslope to the northeast (points 22–26)
(Plate II, upper right). The larger trees in
Laupa
¯
hoehoe are substantially taller than
those at Thurston (>25 m versus <20 m),
and their crowns are broader as well; indi-
vidual trees stand out in the Laupa
¯
hoehoe
canopy-height image (for example, a tall
Acacia koa A. Gray at 32), but they rarely are
distinguishable at Thurston (or the older
sites). At the same time, there are extensive
areas of shorter canopies in the Laupa
¯
hoehoe
landscape—more so than within native forest
at Thurston. The shorter areas reflect human
land use (ongoing koa logging in the adjacent,
partially cleared Waipunalei ahupua‘a [4, 5]
and some past logging in the upper-elevation
portion of Laupa
¯
hoehoe itself ), a few small
hollows and bogs (20), and, most important,
the legacy of canopy dieback (Mueller-
Dombois 1986), especially on the deep volca-
nic ash substrates. Ground observations re-
veal that much of the dieback area has a
canopy of tall Cibotium spp. tree ferns, with
standing dead trees (7, 35).
Canopy N concentrations also vary across
the Laupa
¯
hoehoe landscape (Plate II), with
many value s in excess of those observed in
the Thurston landscape. In part high N con-
centrations result from the presence of Acacia
koa, an N-fixer, within the Laupa
¯
hoehoe for-
est; tall individuals of Metrosideros and Acacia
are readily distinguishable in the imagery
based on their N signature (22, 24), as ground
observations here quickly confirmed. How-
ever, even Metrosideros is substantially higher
in N at Laupa
¯
hoehoe than at Thurston, as
was observed earlier in ground-based sam-
pling (Vitousek et al. 1995). The N image
demonstrates that Acacia reaches its distribu-
tional limit within the Laupa
¯
hoehoe image,
as evidenced by the arc of higher-N canopies
that passes through points 29, 30, and 41; it is
absent from the slightly wetter forests in the
southeastern portion of the image and more
abundant in the drier (and more disturbed)
forests to the northwest. Acacia also responds
strongly to soil scarification, contributing to
its abundance along the road (28, 41) and the
Laupa
¯
hoehoe-Waipunalei boundary fence
line (5) as well as in logged areas.
Kohala Landscape (150 kyr)
Kohala is a smaller mountain than Mauna
Kea, and the focal LSAG site is located just
400 m south-southeast of a block fault (point
10) that bounds a central graben near the
mountain summit (Plate III). The topo-
graphic image demonstrates that volcanic
features are surprisingly prominent in this
landscape, for a region that has not experi-
enced local volcanism for P150 kyr. The
landscape contains several cinder cones that
erupted 150–220 kyr ago and several tra-
chyte lava flows (including a large one west-
northwest of 53) representative of this later
(alkalic) stage of Hawaiian volcanism. The fo-
cal site itself is on an interfluve underlain by
alkalic ‘a‘a
¯
lava; the local drainage netw ork is
coarser and deeper than that in the Laupa
¯
-
hoehoe landscape.
Canopy heights are relatively short and ex-
tremely variable across the Kohala landscape
(Plate III), reflectin g the influence of a steep
rainfall gradient and multiple additional con-
trolling factors. The southern one-third of
the landscape has been cleared and converted
to pasture (15, 32, 33, 53); trees remain in
that landscape primarily in and near gullies.
Several species of exotic trees were planted
in the 1920s and 1930s at and near the upper
margin of the pasture; some of these (Euca-
lyptus spp. and Cryptomeria japonica (L. f.) D.
Don in particular) (14, 18, 31, 35) represent
the tallest trees in this and the older Ko
¯
ke‘e
landscapes.
Aircraft-Based Analysis of Forest Landscapes
.
Vitousek et al. 363

The area beyond the block fault north-
northeast of the focal site (43–52) receives
substantially more rainfall than the LSAG
site, and it supports areas of short-canopy
Metrosideros forest interspersed with Sphag-
num palustre L. and Metrosideros-covered
bogs with occasional patches of taller Metrosi-
deros (43, 45). Finally, the region to the west-
northwest of the focal site is similar to it
climatically, but canopies are short and co-
dominated by the mat-forming fern Dicranop-
teris linearis. Fern dominance reflects dieback
of the Metrosideros overstory in some porti ons
of the area, with standing dead trees over a
nearly monospecific layer of fern (28, 36,
40); in other areas, the Dicranopteris is climb-
ing into what appears to be a growing, post-
disturbance stand of Metrosideros (27, 41).
Canopy N also varies across the landscape:
with the highest concentrations in the tree
plantations and in postdieback Dicranopteris,
and relatively low concentrations in the pas-
ture and most of the Metrosideros forest (Plate
III). Ground observations revealed that sev-
eral areas near the forest-pasture boundary
with moderately tall canopies and high can-
opy N concentrations (22, 34) are tree plan-
tations dominated by N-fixing Casuarina
equisetifolia L. The wet forest region to the
northeast of the focal site includes a fine-
scale matrix of areas with high and low N
concentrations—a surprising result, because
high rainfall generally is associated with lower
N concentrations in Hawaiian forests (Schuur
and Matson 2001). Following up on these re-
mote observations, we found that the higher
N concentrations reflect in part a greater
abundance of the relatively high-N native
tree Cheirodendron trigynum (Gaud) A. Heller
in the canopy in this region (51, 52), as well as
open areas in which the N-rich alien herbs
Hedychium gardnerianum Sheppard ex Ker
Gawl and Tibouchina herbacea (DC) Cogn.
are present in the canopy (50).
Ko
¯
ke‘e Landscape (4,100 kyr)
The Ko
¯
ke‘e area is much older than the other
regions on the LSAG, and more of its topog-
raphy is erosional and depositional (Plate
IV ). However, there is a well-defined area
underlain by a constructional volcanic surface
on which the focal site is located. This area
has a less-pronounced local drainage net-
work (aside from the large and deep valleys
that bound and bisect it) than occurs on the
younger Laupa
¯
hoehoe and Kohala land-
scapes, because it is underlain by thick, flat-
lying caldera-filling deposits of the Olokele
Formation (Sherrod et al. 2007), which are
more resistant to dissection than most Hawai-
ian lavas.
Tree heights generally are short across the
Ko
¯
ke‘e landscape, on both constructional and
erosional/depositional surfaces (Plate IV ).
Major exceptions to this pattern are a tall
Cryptomeria plantation along a stream in the
lower-rainfall southern part of the landscape
(29), other pockets of planted Crytomeria
elsewhere (14, 15), and an arc of taller Metro-
sideros forest 500 m east of the focal site
(23–25). This taller native forest is in a topo-
graphic depression and so probably has richer
soils than the constructional surface (Porder
et al. 2005b); it is also sheltered from strong
winds, and trees there persisted through hur-
ricanes that passed over this landscape in
1957, 1982, and 1992. One striking feature
of the Ko
¯
ke‘e landscape is an open bog in
the northeast (16); this area is covered by
sedges and very short (P0.5 m) Metrosideros,
as is much of the Alaka‘i Swamp to the east.
The areas of higher canopy N are gener-
ally associ ated with nonnative species: in the
planted Cryptomeria and an area of invasive
Morella and Psidium cattleianum Sabine (30,
31) (Plate IV ) in the southern part of the
area, and in smaller areas associated with
Cryptomeria plantings or invasions by N-
fixing Acacia melanoxylon R. Br. (5) and
Morella faya (13). In addition, ground obser-
vations show that a stream valley on the
northwestern side of the image is dominated
by the high-N invasive herb Hedychium
gardnerianum. On the constructional surface,
concentrations of N in Metrosideros gener-
ally are low; concentrations are higher in the
sheltered, Metrosideros-dominated depression
already described. Mats of Dicranopteris
fern, which cov er large portions of the con-
structional surface and many slopes to the
southwest (9, 20, 21, 32), also have low con-
centrations of N. Overall, erosion and depo-
sition influence more of the Ko
¯
ke‘e landscape
364 PACIFIC SCIENCE
.
July 2010

Plate 1. Topography and canopy properties in the 2 km by 2 km landscape centered on the 300-yr-old Thurston
site on a substrate age gradient across the Hawaiian Archipelago. (Upper left) Topography derived from LiDAR. The
long-term research site at Thurston is indicated by a circle with 50 m radius; points where ground-based information
on forest structure and composition and canopy photographs were obtained are indicated by numbers. Images that
allow zooming in are accessible online at http://www.stanford.edu/group/Vitousek/lsaglandscape.htm. (Upper right)
Distribution of canopy heights across the Thurston landscape, also obtained via LiDAR. (Lower left) Nitrogen (N)
concentration in the forest canopy, measured with High-Fidelity Imaging Spectrometer (HiFIS). (Lower right) Cover
classes in the Thurston landscape, derived from the remote images and ground observations. Dark green represents
native-dominated ecosystems on constructional volcanic surfaces; dark blue represents sites with invasive trees (here
predominantly the N-fixer Morella faya) at least codominant in the canopy; yellow represents volcanic features, here
the wall and floor of Kõlauea Iki Crater and some lava-tube skylights; and red represents human-disturbed sites, here
active and abandoned pasture in the northern portion of the image and National Park infrastructure (roads and parking
lots) elsewhere. This figure is revised and reorganized from Vitousek et al. (2009).

Plate 11. Topography and canopy properties in the 2 km by 2 km landscape centered on the 20 kyr Laupëhoehoe
landscape, organized as in Plate I. Cover classes include native-dominated ecosystems on constructional surfaces (dark
green); eroded areas (light green); and human-disturbed areas (red), here mostly logging of Acacia koa in what was once
forested pasture.

Plate 111. Topography and canopy properties in the 2 km by 2 km landscape centered on the 150 kyr Kohala land-
scape, organized as in Plate I. Cover classes include native-dominated ecosystems on constructional surfaces (dark
green); eroded and depositional areas (light green); volcanic and tectonic features (yellow), here cinder cones, the steep
sides of trachyte flows, and a fault scarp; human-disturbed areas (red), here mostly pasture; and plantations of intro-
duced trees (light blue). The hatched area to the northeast receives substantially more rainfall than the remainder of
the image, or the other landscapes.

Plate 1v. Topography and canopy properties in the 2 km by 2 km landscape centered on the 4,100 kyr Kýke‘e land-
scape, organized as in Plate I. Cover classes include native-dominated ecosystems on constructional surfaces (dark
green); eroded and depositional areas (light green); invaded areas (dark blue), here mostly Morella faya and Psidium
cattleianum, with some Hedychium gardnerianum and Acacia melanoxylon; human-disturbed areas (red), here mostly a
road and picnic area; and plantations of introduced trees (light blue).

than they do in any of the younger ones
(Plate IV ). Tree plantations and biological
invasions also contribute to landscape-level
variation in canopy structure and chemistry,
particularly on lower slopes and in valley bot-
toms.
conclusions
Aircraft-based remote sensing can provide
highly resolved, synoptic analyses of the to-
pography, canopy heights, and canopy N
concentrations of forested landscapes; espe-
cially when supplemented by ground-based
observations, they can provide far richer
information than can be obtained at com-
parable spatial scales from ground-based
studies alone. These analyses do not depend
on either extrapolations of known sites or
‘‘ground-truthing’’ of remotely sensed im-
ages; rather they are direct, distributed mea-
surements of ecosystem properties on the
landscape scale. Here we used the remote
measurements to provide a regional perspec-
tive on the results of long-term studies in
particular site s; remote observ ations provide
a direct measure of ecosystem properties
on the landscape scale as these vary with
long-term soil and landscape development.
Aircraft-based measurements can also identify
and analyze ecosystem dynamics that might
be difficult to observe from the ground. Fi-
nally, they can be enjoyable, because on-the-
ground measurements designed to pursue the
sources of variance detected by airborne re-
mote sensing require that investigators get
into every unique portion of a study land-
scape. In doing so, new features of the land-
scape, and new biological communities, may
be discovered. For example, the Ko
¯
ke‘e tree-
height imagery (Plate IV ) suggested to us
that there might be an incipient bog like that
at point 16 developing on an adjacent ridge,
in the area near point 9. Ground-based obser-
vations showed that this area is not boglike;
rather, it is dominated by a Dicranopteris
thicket. However, in getting there we found
striking ‘‘barrens’’ of sparse Metrosideros can-
opies over soil surfaces with pedogenic con-
cretions that looked like a rough young ‘a‘a
¯
lava flow at and around point 8. We had not
observed this combination of features previ-
ously and now are attempting to understand
the factors that brought it into being.
acknowledgments
We thank T. Varga, D. Knapp, T. Kennedy-
Bowdoin, R. Martin, M. Jones, M. Eastwood,
P. Gardner, S. Lundeen, C. Sarture, and R.
Green for airborne data collection and analy-
sis support. Douglas Turner prepared the
Web-based supplementary material.
Literature Cited
Asner, G. P., R. F. Hughes, T. A. Varga, D. E.
Knapp, and T. Kennedy-Bowdoin. 2009.
Environmental and biotic controls on
aboveground biomass throughout a Ha-
waiian rainforest. Ecosystems 12:261–278.
Asner, G. P., R. F. Hughes, P. M. Vitousek,
D. E. Knapp, T. Kennedy-Bowdoin, J.
Boardman, R. E. Martin, M. Eastwood,
and R. O. Green. 2008. Invasive plants
alter 3-D structure of rainforests. Proc.
Natl. Acad. Sci. U.S.A. 105:4519–4523.
Asner, G. P., D. E. Knapp, M. O. Jones,
T. Kennedy-Bowdoin, R. E. Martin, J.
Boardman, and C. B. Field. 2007. Carne-
gie Airborne Observatory: In-flight fusion
of hyperspectral imaging and waveform
light detection and ranging (wLiDAR) for
three-dimensional studies of ecosystems.
J. Appl. Remote Sens. 1: doi: 10.1117/
1.2794018.
Asner, G. P., and P. M. Vitousek. 2005. Re-
mote analysis of biological invasion and
biogeochemical change. Proc. Natl. Acad.
Sci. U.S.A. 102:4383–4386.
Chadwick, O. A., L. Derry, P. M. Vitousek,
B. J. Huebert, and L. O. Hedin. 1999.
Changing sources of nutrients during four
million years of ecosystem development.
Nature (Lond.) 397:491–497.
Crews, T. E., K. Kitayama, J. Fownes, D.
Herbert, D. Mueller-Dombois, R. H.
Riley, and P. M. Vitousek. 1995. Changes in
soil phosphorus and ecosyst em dynamics
across a long soil chronosequence in Ha-
wai‘i. Ecology 76:1407–1424.
Green, R. O., M. L. Eastwood, C. M. Sar-
ture, T. G. Chrien, M. Aronsson, B. J.
Aircraft-Based Analysis of Forest Landscapes
.
Vitousek et al. 365

Chippendale, J. A. Faust, B. E. Pavri, C. J.
Chovit, and M. Solis. 1998. Imaging spec-
troscopy and the Airborne Visible/Infrared
Imaging Spectrometer (AVIRIS). Remote
Sens. Environ. 65:227–248.
Harrington, R. A., J. H. Fownes, and P. M.
Vitousek. 2001. Production and resource-
use efficiencies in N- and P-limited tropi-
cal forest ecosystems. Ecosystems 4:646–
657.
Herbert, D. A., and J. H. Fownes. 1999. For-
est productivity and efficiency of resource
use across a chronosequence of tropical
montane soils. Ecosystems 2:242–254.
Hobbie, S. E., and P. M. Vitousek. 2000. Nu-
trient limitation of decomposition in Ha-
waiian forests. Ecology 81:1867–1877.
Kitayama, K., and D. Mueller-Dombois.
1995. Vegetation changes along gradients
of long-term soil development in the Ha-
waiian montane rain forest zone. Vegetatio
120:1–20.
Kupiec, J. A., and P. J. Curran. 1995. De-
coupling effects of the canopy and foliar
biochemicals in AVIRIS spe ctra. Int. J. Re-
mote Sens. 16:1731–1739.
Lefsky, M. A., W. B. Cohen, G. G. Parker,
and D. J. Harding. 2002. Lidar remote
sensing for ecosystem studies. BioScience
52:19–30.
Mueller-Dombois, D. 1986. Perspectives for
an etiology of stand-level dieback. Annu.
Rev. Ecol. Syst. 17:221–243.
Porder, S. J., G. P. Asner, and P. M. Vitou-
sek. 2005a. Ground-based and remotely-
sensed nutrient availability across a tropi-
cal landscape. Proc. Natl. Acad. Sci.
U.S.A. 102:10909–10912.
Porder, S. J., A. Paytan, and P. M. Vitousek.
2005b. Erosion and landscape development
affect plant nutrient status in the Hawaiian
Islands. Oecologia (Berl.) 142:440–449.
Riley, R. H., and P. M. Vitousek. 1995. Nu-
trient dynamics and trace gas flux during
ecosystem deve lopment in Hawaiian mon-
tane rainforest. Ecology 76:292–304.
Roberts, D. A., M. O. Smith, and J. B.
Adams. 1993. Green vegetation, non-
photosynthetic vegetation, and soils in
AVIRIS data. Remote Sens. Environ.
44:255–269.
Schuur, E. A. G., and P. A. Matson. 2001.
Aboveground net primary productivity
and nutrient cycling across a mesic to wet
precipitation gradient in Hawaiian mon-
tane forest. Oecologia (Berl.) 128:431–
442.
Sherrod, D. R., J. M. Sinton, S. E. Watkins,
and K. M. Brunt. 2007. Geologic map of
the state of Hawai‘i. U.S. Geol. Surv.
Open File Rep. 2007-1089.
Smith, M.-L., S. V. Ollinger, M. E. Martin,
J. D. Aber, R. A. Hallett, and C. L. Good-
ale. 2002. Dir ect estimation of above-
ground forest productivity through hy-
perspectral remote sensing of canopy
nitrogen. Ecol. Appl. 12:1286–1302.
Su, J., and E. Bork. 2006. Influence of vege-
tation, slope and lidar sampling angle on
DEM accuracy. Photogram. Engin. Re-
mote Sens. 72:1265–1274.
Ustin, S. L., D. A. Roberts, J. A. Gamon ,
G. P. Asner, and R. O. Green. 2004. Using
imaging spectroscopy to study ecosystem
processes and properties. BioScience
54:523–534.
Vitousek, P. M. 2004. Nutrient cycling and
limitation: Hawai‘i as a model system.
Princeton University Press, Princeton,
New Jersey.
Vitousek, P. M., G. P. Asner, O. A.
Chadwick, and S. C. Hotchkiss. 2009.
Landscape-level variation in forest struc-
ture and biogeochemistry across a sub-
strate age gradient in Hawai‘i. Ecology
90:3074–3086.
Vitousek, P. M., and H . Farrington. 1997.
Nutrient limitation and soil development:
Experimental test of a biogeochemical
theory. Biogeochemistry (Dordr.) 37:63–
75.
Vitousek, P. M., D. R. Turner, and K.
Kitayama. 1995. Foliar nutrients during
long-term soil development in Hawaiian
montane rain forest. Ecology 76:712–720.
Wiegand, B. A., O. A. Chadwick, P. M.
Vitousek, and J. L. Wooden. 2005. Ca
cycling and isotopic fluxes in forested eco-
systems in Hawaii. Geophys. Res. Lett.
32:L11404, doi: 10.1029/2005GL022746.
Wolfe, E. W., and J. Morris. 1996. Geologic
map of the island of Hawaii. U.S. Geolog-
ical Survey Map 1-2524-A, Miscellaneous
Investigations Series, Reston, Virginia.
366 PACIFIC SCIENCE
.
July 2010