l t
rn
b,
euve
CA 9
serv
Received in revised form 2 April 2015
Accepted 8 April 2015
Available online xxxx
land-cover types. An understanding of how tropical forests respond to
cape scale has thus fo-
omy, relatively small
amily-level taxonomy
nsing was most useful
Remote Sensing of Environment xxx (2015) xxx–xxx
RSE-09394; No of Pages 10
Contents lists available at ScienceDirect
Remote Sensing o
.ehow the composition, structure and function of tropical forest canopies
will respond to changing environmental conditions will increase as the
rate of change accelerates (Schimel, Asner, & Moorcroft, 2013).
The evidence for a pantropical response to global anthropogenic
forcing comes almost exclusively from relatively small-scaled censuses
(Asner, 2013). The majority of work at the lands
cused on general description of forest physiogn
spatial domains, subsets of common species, or f
(e.g. Higgins et al., 2014).
Until recently, air- and spaceborne remote seAnticipated changes in regional and global climate could drive shifts in
the geographic extent, composition and condition of tropical forest cano-
pies (Collwell, Brehm, Cardelus, Gilman, & Longino, 2008; Wright, 2005).
Biologists, conservationists and policy makers therefore raise concerns
about alterations in the functioning of tropical forests and their capacity
to sustain environmental services such as carbon storage and water provi-
sioning (FAO, 2007; Foster, 2001). A need for thorough understanding of
environmental change requires scaling up our observation capability
to the landscape level that captures entire forest communities and tran-
sitions between communities. Yet, our ability to measure, scale up and
predict basic ecosystem function in tropical forests remains weak. This
is strongly linked to practical and logistic difficulties in the often inac-
cessible tall forest canopies and the overwhelming local-scale (alpha)
and regional-scale (beta, gamma) diversity of many tropical systemsof tree plots (Wright, 2005). Although these n
provide valuable insights to the fundamental p
opy function, they lack scalability due to the e
of tropical canopies in terms of both floristic
⁎ Corresponding author.
E-mail address: ben.somers@ees.kuleuven.be (B. Some
http://dx.doi.org/10.1016/j.rse.2015.04.016
0034-4257/© 2015 Elsevier Inc. All rights reserved.
Please cite this article as: Somers, B., et al., M
Panama using airborne imaging s..., Remote Sas well as their non-random or systematic placement across tropical1. IntroductionKeywords:
Biodiversity
Hyperspectral
Climate change
Spectral variation hypothesis
Alpha diversity
Beta diversitya bioclimatic gradient in Panama. The expressed precipitation gradient from the wet Caribbean side to the dry
Pacific side makes Panama an excellent study area for performing a mesoscale assessment of climate effects on
tropical tree species richness. Spatial patterns in local spectral variability (expressed as the coefficient of varia-
tion) and spectral similarity (expressed as the spectral similarity index) were used as proxies for species area
curves and species distance decay curves. Our analysis revealed significant spectral changes along the precipita-
tion gradient. Highest spectral diversity was observed for moist forest sites while lowest diversity was observed
for the driest forest sites. Most of the spectral variation came from changes in the visible (VIS) and shortwave-
infrared (SWIR) reflectance. Variation in the VIS was significantly higher for the dry compared to the moist
and wet forests, while the opposite was true for the NIR and SWIR reflectance. Our spectral mesoscale analysis
extends previous results suggesting that niche differentiationwith respect to soil water availability is a direct de-
terminant of both local- and regional-scale distributions of tropical trees. A next step would be to test the accu-
racy and scalability of our results with lower spatial resolution spectrometer data, simulating the observing
conditions that will be achieved with future satellite missions such as the European Union's EnMap and NASA's
HyspIRI missions.
© 2015 Elsevier Inc. All rights reserved.Article history:
Received 15 June 2014
We used imaging spectroscopy to perform a top-downmesoscale analysis of tropical tree species richness acrossa b s t r a c ta r t i c l e i n f oMesoscale assessment of changes in tropica
bioclimatic gradient in Panama using airbo
Ben Somers a,d,⁎, Gregory P. Asner b, Roberta E. Martin
S. Joseph Wright c, Ruben Van De Kerchove d
a Division Forest, Nature & Landscape, KU Leuven, Celestijnenlaan 200E — bus 2411, B-3001 L
b Department of Global Ecology, Carnegie Institution for Science, 260 Panama Street, Stanford,
c Smithsonian Tropical Research Institute, Balbao, Ancon, Panama
d Flemish Institute for Technological Research (VITO), Centre for Remote Sensing and Earth Ob
j ourna l homepage: wwwetworks of observations
rocesses governing can-
xtremely diverse nature
and structural variation,
rs).
esoscale assessment of cha
ensing of Environment (2015)ree species richness across a
e imaging spectroscopy
Christopher B. Anderson b, David E. Knapp b,
n, Belgium
4305, USA
ation Processes (TAP), Boeretang 200, BE-2400 Mol, Belgium
f Environment
l sev ie r .com/ locate / rsefor determining the spatial extent and dynamics of vegetation cover.
However, technical developments in sensors and instrumentation
have vastly improved the quantity and quality of information that can
be obtained remotely, and advances in understanding how light inter-
acts with plant canopies have made remote sensing increasingly useful
for detecting patterns and analyzing processes related to the composi-
tion and functioning of vegetated ecosystems. Imaging spectroscopy, a
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
2 B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxremote sensing technology capable of measuring the earth's reflectance
as a continuous spectrum of dozens to hundreds of narrow spectral
bands across the visible and near-infrared spectral domain, has shown
great potential to map the structure, function and composition of eco-
systems at the “mesoscale” (e.g. Jusoff & Ibrahim, 2009; Ustin, Roberts,
Gamon, Asner, & Green, 2004). The measured reflectance spectra are
sensitive to the structural organization of, and variations in chemical
constituents in, canopy components. These physico-chemical-to-
spectral linkages provide ameans of detecting species and/or functional
types (e.g., Asner & Martin, 2009; Asner & Vitousek, 2005; Clark,
Roberts, & Clark, 2005; Somers & Asner, 2012; Ustin & Gamon, 2010),
and can even provide information about the biogeochemical heteroge-
neity (e.g. Townsend, Asner, & Cleveland, 2008; Vitousek, Asner,
Chadwick, & Hotchkiss, 2009) and species richness of tropical forest
canopies (e.g., Asner, Nepstad, Cardinot, & Ray, 2004; Carlson, Asner,
Hughes, Ostertag, & Martin, 2007; Feret & Asner, 2013; Kalacska et al.,
2007; Nagendra & Rocchini, 2008; Somers and Asner, 2013).
The remotemapping of biological and/or functional diversity is often
done by analyzing variation of a particular spectral signal or spectral fea-
ture (Gould, 2000). This Spectral Variation Hypothesis (SVH) relies on the
positive relationship between biological diversity and environmental
heterogeneity, and has been used tomap or detect biodiversity hotspots
(alpha-diversity) and species turnover (beta-diversity) within and be-
tween a variety of ecosystems and communities (e.g., Gillespie, Foody,
Rocchini, Giorgi, & Saatchi, 2008; Nagendra & Rocchini, 2008; Baldeck
& Asner, 2013). Despite progress in the use of spectral variation to
estimate biological diversity at different ecological scales, we still lack
approaches needed to yield consistent and comparable biodiversity in-
formation across different ecosystems. This is particularly true in tropi-
cal regions where, for example, vegetation communities may vary from
dry to humid forests often over short distances due to strong regional
climate gradients (Condit, Ashton, Bunyavejchewin, et al., 2006). With
global climate change, it is expected that the current environmental gra-
dients under which forest assemblages formed may shift, and plant
communities will be altered in response to those shifts. However, the
extent, pattern, and rate of change in forest composition remain un-
known, and most dynamic vegetation models lack the fine-scale geo-
graphic and biological resolution needed to predict plant community
changes over time (Schimel et al., 2013).
New methods and technologies are critically needed to map and
monitor changes in the functional and biological composition of ecosys-
tems through time. Nowhere does this seem more critical than in trop-
ical regions, such as Panama, where climate change and land use come
together to place maximum pressure on forests and the ecological ser-
vices they provide to society. Herewe use airborne imaging spectrosco-
py to perform a top-downmesoscale analysis of changes in tropical tree
species richness across a bioclimatic gradient in Panama. The expressed
precipitation gradient from the wet Caribbean side to the dry Pacific
side makes Panama an excellent study area. We sought to answer
these specific questions: (i) Can we use airborne imaging spectroscopy
to study spatial patterns in local (alpha) and regional (beta) tree species
richness across tropical forests? (ii) Canwe reveal significant changes in
forest canopy spectral patterns, and thus canopy composition and di-
versity, along a precipitation gradient in Panama?; and if so (iii) are
there specific spectral regions or wavelengths that dominate the spec-
tral variation? In this studywe seek to determine if imaging spectrosco-
py can be used to scale up previous results from plot-based studies
providing a technology to track shifts in species richness due to climate
change over broad spatial scales.
2. Material
2.1. Study area
The isthmus of Panama is dominated by a strong environmental gra-
dient in climate, topography and geology. Average annual precipitation
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)ranges from less than 1600 mm/yr on the Pacific side of the isthmus
gradually increasing to over 3100 mm/yr on the Caribbean coast. At
the highest elevations along the Caribbean coast precipitation can
reach 4000 mm/yr (Rand & Rand, 1982). Rainfall is seasonal with a
dry season from January through March, showing marked variation
across sites, with an annual extreme moisture deficit around 500–
600 mm at the driest sites but only between 300 and 400 mm in the
wettest sides (Condit, Engelbrecht, Pino, Pérez, & Turner, 2013). The
weathering pattern produced by the strong precipitation gradient has
resulted in a complex geological terrain composed of either dense, rel-
atively impermeable volcanic rock or porous, chemically unstable sedi-
mentary rocks and volcanic mud flow deposits (Dietrich, Windsor, &
Dunne, 1982).
Due to the variation in rainfall, Panama harbors a great diversity of
tree species. The isthmus can broadly be divided into three general biocli-
matic regions. On thewettest Caribbean slopes, there is enoughmoisture
throughout the year to support evergreen tropical forests. In contrast, on
the Pacific side many of the slopes have hard, dry soil by April. On this
south-western side, many species are dry-season deciduous. In the mid-
dle of the country, lies moist tropical forest where the community transi-
tions from dry to wet along the precipitation gradient. The trees increase
in size and the occurrence of deciduousness lessens compared to dry for-
ests, but does not disappear entirely (Condit, Pérez, & Daguerre, 2010).
We selected a representative site of approximately 400 ha in each of
the three bioclimatic regions (dry forest site: 7°26′50″N, 80°10′45″W;
moist forest site: 9°4′32″N, 79°39′12″W; wet forest site: 9°16′50″N,
79°58′44″W)where both airborne imagery (see Section 2.3) and ground
reference data (see Section 2.2) were available (Fig. 1).
2.2. Floristic data
For this study we used publicly available species lists collected from
18 permanent sampling forest plots (10 plots of 1 ha and 8 plots of
0.4 ha, Table 1) maintained by the Smithsonian Institution's Center for
Tropical Forest Science (Condit, 1998; Pyke, Condit, Salamon, & Lao,
2001). For each plot all tree stems ≥ 10 cm DBH were identified and
listed. These data were used to validate the spectral proxies for species
richness and turnover (cf. Section 3.1.).
2.3. Remote sensing data and preprocessing
For each of the three study sites (Fig. 1) we used data collected from
the Carnegie Airborne Observatory-2 Airborne Taxonomic Mapping
Systems (CAO-2 AToMS; Asner et al., 2012). The imagery was acquired
during January–February 2012 (i.e. the early dry season). AToMS in-
cludes a Visible-to-ShortWave InfraRed (VSWIR) imaging spectrometer
and a dual laser, waveform LiDAR (Asner et al., 2012). These sub-
systems are boresight aligned onboard a Dornier 228-202 aircraft.
Datawere collected from an altitude of 2000m above ground level, pro-
viding imagery with a 2 m spatial resolution, at an average flight speed
of 55–60 m s−1 and a mapping swath of 1.2 km.
The VSWIR spectrometer collects data in 480 contiguous spectral
bands spanning the 252–2648 nm wavelength range with a spectral
resolution of 5 nm. The VSWIR data were radiometrically corrected
using a flat-field correction, radiometric calibration coefficients, and
spectral calibration data collected in the laboratory. Apparent surface
reflectancewas derived from the radiance values using the ACORN-5 at-
mospheric correction model (Imspec LLC, Palmdale, CA). To improve
aerosol corrections, ACORN-5 was run iteratively with different visibili-
ties until the reflectance at 420 nm (i.e. almost constant for vegetation)
was 1%. The reflectance data were further corrected for cross-track
brightness gradients using a bidirectional reflectance distribution func-
tion model (Colgan, Baldeck, Féret, & Asner, 2012). Full details on the
preprocessing of the VSWIR data can be found in Asner et al. (2014)
and Colgan et al. (2012).
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
Wet forest
3B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxDry forestThe LiDAR in the CAO-2AToMS is a dual-laser scanning systemoper-
ating at 1064 nm. The LiDAR collects the full waveform with up to four
discrete returns per laser shot. The LiDAR sub-system was configured
such that the laser point density achieved was approximately 2 shots
per square meter (or 8 shots per VSWIR pixel). From the LiDAR point
cloud data, a physically-based model was used to estimate top-of-
canopy and ground surfaces using Terrascan/Terramatch (Terasolid
Ltd., Jyväskylä, Finland) software packages. Vegetation height was
then estimated by differencing the top-of-canopy and ground surface
digital elevation models following the common approach for these
data (e.g. Lefsky et al., 1999). These structural data allowed for automat-
ed masking of forest gaps, intra- and inter-canopy shadows, and mini-
mum vegetation height in the VSWIR images (Asner et al., 2008). A
minimum LiDAR vegetation height requirement of 5 m was applied to
remove exposed ground areas and non-tree vegetation. The LiDAR ras-
ter resolution was 1 m. In addition, a NDVI mask was applied (all pixels
with a NDVI value b 0.4 were ignored) to exclude all remaining
nonphotosynthetically active vegetation (NVP) and man-made mate-
rials (e.g. buildings). Subsequently, clouds and cloud shadows were
manually masked. The remaining sunlit canopy spectra were used to
analyze spatial patterns in forest canopy diversity and composition.
Fig. 1. Panamanian isthmus and loc
Table 1
Basic summary information on number (and size) of plots per forest type and response variab
0.4 ha plots
Number of plots Average species richness (and standard devia
Dry forest site 1 35 (NA)
Moist forest site 7 78 (16)
Wet forest site 2 73 (5)
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)Moist forest3. Methods
3.1. Validation of remotely sensed proxies of tree species richness
Research on plant species diversity and abundance mapping using
remote sensing are broadly based on the Spectral Variation Hypothesis
(SVH; e.g. Palmer, Earls, Hoagland, White, & Wohlgemuth, 2002;
Rocchini, Balkenhol, Carter, et al., 2010) which relies on the assumption
that spectral heterogeneity can be used to quantify (species) diversity.
The obtained species diversity patterns are in turnbelieved to also be re-
lated to environmental (ecosystem) heterogeneity, based on the ‘port-
folio effect’ (Rocchini et al., 2010). The SVH has been mainly used for
two purposes: (1) the mapping or detection of biodiversity hotspots
(so-called α-diversity); and (2) the development of quantitative mea-
sures for species turnover between ecosystems (so-called β-diversity).
Several recent studies have, indeed, verified that local spectral vari-
ability in remote sensing data correlates with local plant species rich-
ness (i.e. the number of species per unit area; e.g. Rocchini, 2007) in a
variety of ecosystems. Many of these studies showed that measures of
dispersion, such as the coefficient of variation (CV), are simple and ef-
fective indicators of spectral heterogeneity per sampling unit (e.g. an
ations of the three study sites.
le (species richness).
1 ha plots
tion) Number of plots Average species richness (and standard deviation)
1 59 (NA)
5 80 (11)
2 80 (2)
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
composition changes among sites. An important approach to measure
spatial variation in beta diversity or species turnover is the distance-
decay of community similarity. Distance-decay studies regress pair-
wise measures of sample-unit similarity against pair-wise spatial dis-
tance, and parameterize a ‘slope’ that indicates the relative change in
compositional similarity through geographic space (Morlon et al.,
2008). Through calculating the spectral similarity between plots (quan-
tified as SI, Eq. 2) at different spatial distanceswe could generate a spec-
tral proxy for the Distance-Decay curve. These spectral distance-decay
curves allowed us to assess spatial patterns in species turnover or beta
diversity.
For each of the three study sites the spectral species area and dis-
tance decay curves were calculated for each location (i.e. each image
pixel). This is done using a moving window approach. For each pixel
CV and SI were calculated for a series of image windows, which were
square kernels of 3 × 3 to 45 × 45 pixels. This allowed reconstruction
of the species area and distance decay curves for each individual
image pixel. The spectral proxies were first averaged over the entire
spectrum and also calculated for each waveband separately. The spec-
tral proxies, calculated using individual bands or the whole spectrum,
and relationships derived from them were combined to assess spatial
patterns in species diversity along the bioclimatic gradient. Statistical
analysis was performed using version 3.0.2 of the 64-bit version of R, a
multi-platform, open-source language and software for statistical com-
puting (R Development Core Team, 2010). All statistical analyses were
evaluated against the 95% confidence interval.
4 B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxximage window/kernel) (e.g. Carter, Knapp, Anderson, Hoch, & Smith,
2005; Duro et al., 2014; Levin, Shmida, Levanoni, Tamari, & Kark,
2007; Lucas & Carter, 2008). Here we employed the CV (Eq. 1) to link
the spectral heterogeneity within the 1 ha or 0.4 ha area of the field
plots (equivalent to a sampling window of 50 by 50 (1 ha) and 32 by
32 (0.4 ha) image pixels respectively), to the field species counts within
the field plots (cf. Section 2.2).
CV ¼
1
n
Xn
b¼0
sd Rbð Þ
mean Rbð Þ
ð1Þ
WithRb the reflectance in bandb and n the total number of bands. CV
can either be calculated for the full spectrum, a part of the spectrum or
on a per waveband basis. In the latter case n equals 1. CV increases with
increasing spectral variability and is, according the SVH hypothesis, as
such, a measure for species richness, a common measure of alpha-
diversity (Rocchini, Dadalt, Delucchi, Neteler, & Palmer, 2014, Rocchini
et al., 2013). Since CV is a bounded variable, regression analysiswas per-
formed using a General Linear Model with a Gamma error distribution.
The SVH further suggests that beta-diversity or species turnover can
be quantified using the spectral distance (i.e. the spectral similarity) be-
tween different locations (or image pixel windows). The rationale is
that, the more similar the spectral populations of two image pixel win-
dows are, the more similar the species pools in both locations. In con-
trast, the larger the spectral distance between both populations, the
more likely the species turnover is larger between the locations
(Rocchini, Butini, & Chiarucci, 2005). Here we use the spectral similarity
index (Eq. 2; Somers, Delalieux, Stuckens, Verstraeten, & Coppin, 2009,
Somers, Delalieux, et al., 2010; Somers & Asner, 2012) to quantify the
spectral overlap between two plots i and j across all study sites:
SI ¼
1
n
Xn
b¼0
sd Rb;i
 
þ sd Rb; j
 
Rb;i−Rb; j




: ð2Þ
SI provides a straightforward way to calculate the spectral distance
between two populations for the full spectrum, a part of the spectrum
or on a per waveband basis (in this case n equals 1). Smaller SI values
were expected to correspond to smaller spectral similarity and thus
higher beta-diversity (smaller species overlap).
Once calibrated, these spectral diversity measures permit a
metascale assessment of the canopy composition and diversity in our
study sites. We evaluated CV (Eq. 1) as an indicator for species richness
at the local/site scale (high alpha-diversity), and used SI to provide in-
formation on differences among sites in terms of turnover in species
composition (beta-diversity).
3.2. Mesoscale assessment of tree species richness along a bioclimatic
gradient
3.2.1. Local diversity patterns through species area curves
In order to evaluate the differences in local diversity patterns be-
tween the three bioclimatic regions, we created spectral proxies for
the relationship between species richness and area by calculating CV
(i.e. a proxy for alpha diversity) for different image kernel/window
sizes. Species-area relationships (SARs) or species area curves (SACs),
measure how the number of observed species increaseswith increasing
sample area, and constitute one of the most important and robust tools
to characterize patterns in local diversity (Gotelli & Colwell, 2001).
Therefore the spectral proxies of the SACs are an essential tool to test
our hypothesis that climate processes are responsible for patterns of
local diversity. In addition, SACs facilitate comparisons ofmeasurements
at different spatial scales.
3.2.2. Patterns in species turnover through distance decay curves
Species area curves give an idea of the rate of change in species rich-ness but do not give insight in how and at which rate species
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)4. Results
4.1. Validation of remotely sensed proxies of tree species richness
In line with previous reports (Carter et al., 2005; Lucas & Carter,
2008; Rocchini et al., 2013, 2014), the spectral variation, quantified as
CV averaged over all bands, showed a significant (p b 0.001) positive
correlation with species richness or alpha diversity in our study area
(residual deviance = 1.01; Fig. 2).
Also corresponding to previous results, the spectral similarity
between plots proved to be a reliable proxy for species turnover or
beta diversity (Fig. 3). Using SI, we successfully modeled the pairwise
Fig. 2. Scatterplot showing the number of species observed in the 18 sampling forest plots
(Section 2.2) against the coefficient of spectral variation averaged over all wavebands. A
GLM model with a gamma error distribution showed a significant positive correlation
between both variables (p-value b 0.001 and residual deviance = 1.01).
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
Fig. 3. Pairwise comparison of the spectral overlap between different plots (the SI index
averaged over all wavebands) and the percentage of species that are common within
5B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxcomparisons of species overlap (R2 = 0.47; p b 0.001). These results
highlight the feasibility of using spectral proxies to perform amesoscale
assessment of changes in tree species richness along the bioclimatic gra-
dient in our study area.
4.2. Mesoscale assessment of tree species richness along a climate gradient
4.2.1. Local diversity patterns through species area curves
For each of the three sites, the average spectral variability–area
curves (see 3.2. and Fig. 4), which can be considered variograms, reflect
typical species-area relationships, with a near linear increase in spectral
variability (species number) with area at smaller spatial scales which
becomes shallower with increasing spatial extent until it finally pla-
teaus (Scheiner, 2003). For all three sites maximal spectral variability
(proxy for species richness), quantified as the CV averaged over the
the sampling forest plots.full spectrum, was reached at an image window size of 21 by 21 pixels
(corresponding to a ground area of 42 by 42 m or approximately
Fig. 4. Average CV as a function of kernel size (i.e. spectral proxy for species area curves)
and 95% confidence interval for the three different study sites. Note that we used squared
kernels.
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)0.18 ha). Yet, a Mann–Whitney–Wilcoxon Test revealed significant dif-
ferences (p-value b 0.001) in the total area under the CV curve (Fig. 4)
among the different sites. The test confirmed a distinctively higher spec-
tral variability for the moist forest site at all spatial scales compared to
both wet and dry forests (about 15% higher). Also, significantly higher
spectral variationwas observed for thewet compared to the dry forests,
yet the main difference between both sites was the clearly smaller var-
iation in the distribution of CV values for thewet site as displayed in the
maps of Fig. 5 showing the spatial variation in the total area under the
spectral species area curve. It is clear that most of the wet forest area
is characterized by a stable spectral variability–area relationship (sum
of CV for individual pixels averaged over all n bands ranges between 1
and 1.4, green/yellow) while only a limited area of low (sum of CV
below 1, dark green) and high spectral diversity (sum of CV above 1.4,
orange/red) are present. In the dry site we see similar patterns with
more areas with a low index value (sum of CV below 1, dark green)
and fewer areas with high index values (sum of CV above 1.6, red).
The ecosystem with the highest alpha diversity was the moist forest
with many large areas with index values exceeding 1.6 (Fig. 5, red).
Calculating the spectral variability–area curves on a per wavelength
basis, as shown in Fig. 6, revealed subtle yet significant differences in
spectral properties, and by correlation canopy chemical composition,
among the different sites. For all three sites the spectral variability in
the near-infrared (NIR, 700–1400 nm) was moderate (CV around 0.2)
when compared to the visible (VIS, 350–700 nm) and shortwave-
infrared (SWIR, 1400–2500 nm) (CV up to 0.45; Fig. 6). Most of the
spectral variation in the VIS region was observed in the dry sites
(CVdry up to 0.35 vs 0.33–0.34 for moist and wet sites respectively,
Fig. 6). A Mann–Whitney–Wilcoxon test, indeed revealed significantly
(p-value = 0.015) higher values for the total area under the CV curve
in the VIS domain for the dry compared to the wet and moist sites. In
contrast, moist and wet sites displayed significantly (p-value b 0.001)
higher variability across the full spectrum and in the SWIR region
(Fig. 6). The VIS (p-value = 0.009) and SWIR (p-value b0.001) reflec-
tance of the dry sites was significantly higher than that of the two
other sites (left panel of Fig. 7) reflecting lower levels of canopy water
content (water absorbs SWIR reflectance) and canopy chlorophyll (pig-
ments strongly absorb light in the visible spectrum) and/or more expo-
sure of bark spectral properties (i.e. bark having high VIS and SWIR
reflectance) during the dry season (Clark & Roberts, 2012).
4.2.2. Patterns in species turnover through distance decay curves
To fully assess and understand the effects of climate on tropical tree
species richness, we need information on species turnover as well. Re-
call from Section 4.1. that the spectral similarity between different
plots (quantified as SI) showed a significant positive relation to species
turnover. A general assessment of the spectral similarity, shown in
Fig. 7, revealed relatively high spectral differences between dry and
wet areas, moderate differences between dry and moist, and small dif-
ferences between moist and wet areas. The small differences between
moist and wet areas were apparent for the full spectrum (Fig. 7). How-
ever, the wet forest showed a slightly lower reflectance in the VIS com-
pared to both other sites (Fig. 7). This can partly be assigned to more
photosynthetically active radiation that is absorbed and thus foliar
pigment concentrations that are higher. Differences in NIR reflectance
between the dry site and the wet and moist sites were relatively small
(SI N 15).
The patterns in species turnover (or spectral similarity index, SI), oc-
curring among locations within each of the bioclimatic regions, indicate
subtle differences as shown in Fig. 8. Slightly lower spectral similarity at
all spatial scales was observed for the dry forest compared to the moist
forestwhich in turn showed slightly less similarity compared to thewet
forest. The precipitation signal is especially expressed in the VIS andNIR.
The similarity in these spectral domains is clearly higher for the wet
compared to themoist site (Fig. 9). In the dry sites we noticed that spec-
tral differences were most pronounced at forest edges as illustrated in
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
6 B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxFig. 10 showing amapof the total area under the spectral distance decay
curve (Fig. 8).
5. Discussion and conclusions
We used airborne imaging spectroscopy as spectral proxies for local
(alpha) and regional (beta) diversity, and revealed significant changes
in spectral properties along a precipitation gradient in Panama. Spatial
patterns in local spectral variability and spectral similarity were used
as proxies for species area curves and species distance decay curves.
The spectral species area curves revealed a lower spectral diversity
for the dry forest at all spatial scales as compared to the wet and moist
forest sites (Figs. 4 and 5). This corroborates previous results of among
others Gentry (1988) who showed that species richness increases with
rainfall in Neotropical forests and reaches an asymptote at about 4man-
nual rainfall. The number of species adapted to the severe seasonal dry
conditions is limited, resulting in an overall lower species diversity
(Engelbrecht et al., 2007; Gentry, 1988). More remarkable, however,
we found distinctly higher spectral variability throughout themoist for-
est compared to the wet forest (Figs. 4 and 5). We found that, on aver-
age, the spectral variability (expressed as total area under the spectral
Fig. 5. (top) Map of the total area under the spectral proxy for SACs (i.e. change in CVwith kern
averaged over all n bands. Dark green areas represent values of total area under the CV curve ra
1.4–1.6, and red between 1.6–1.8. Black rectangles in the left andmiddle panel indicate cloud an
ing hillshaded DEM derived from the LiDAR data. (For interpretation of the references to color
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)proxy for SACs (i.e. change in CV with kernel size, Section 3.2.); Figs. 4
and 5) is significantly higher (about 15%) for the moist compared to
the wet forest site. This tendency towards higher spectral variability in
the moist site is also nicely illustrated in the maps of Fig. 5. Our results
as such suggest an intermediate peak in tree species richness in the
moist forest as compared to the dry or wet forests in our study area.
This partly contradicts earlier reports of Pyke et al. (2001) who found
a greater tree species richness on the wetter side of the Panamanian
isthmus. Yet, in our subsample of the different forest types, the moist
forest plots were slightly richer in species compared to the wet forest
(i.e. average of 73 compared to 78 species/0.4 ha plot in wet compared
to the moist plots respectively; in the 1 ha plots on average approxi-
mately 80 tree species were observed in both the moist and wet forest
sites; here we need to note that only four samples were available for
the wet forests so that these numbers are mainly indicative rather
than providing significant proof of the differences between the sites,
Table 1).
An additional explanation can be the increased occurrence of lianas
in the moist forests. Schnitzer (2005) developed the hypothesis that li-
anas reach peak species richness at intermediate rainfall. Our results
could as such suggest that trees plus canopy lianas reach peak species
el size, Section 3.2) for the three study sites. CV is calculated here for individual pixels and
nging between 0.8–1, light green between 1–1.2, yellow between 1.2–1.4, orange between
d cloud shadows that weremanuallymasked from the analysis; (bottom) the correspond-
in this figure legend, the reader is referred to the web version of this article.)
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
a fu
7B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxrichness in the intermediate, moist forest. Increased epiphyte load and
epiphylls on mature leaves might further add to the higher spectral di-
versity observed in the moist forest (Clark & Clark, 1990).
Another consideration is that themoist forests have greater variabil-
ity in topography, which can be seen from the hill shaded DEMs in the
bottom panels of Fig. 5. Topography affects cloud cover (insolation),
precipitation, and temperature variability. This increased variation in
environmental conditions, which may be more pronounced in the dry
season when imagery were acquired, strongly drives (variation in)
plant species composition (Pau et al., 2013). Comparison of the hill-
shaded DEMs and the CV and SI index maps shown in Figs. 5 and 10,
suggests a tendency towards higher CV and SI values around topograph-
ic transition zones for the moist forest site, a relationship which is less
Fig. 6. Average spectral variability per wavelength (expressed as CV) for each site asexpressed in the wet forest site. This likely prominent topographic con-
trol on spectral diversity in the moist forests might further explain the
observed higher spectral diversity in these sites.
A detailed analysis of the spectral variability on a per wavelength
basis (Fig. 6) further indicated that most of the spectral variation
comes from changes in the VIS and SWIR reflectance. Most interesting
was the observation that the variation in the VIS was significantly
higher for the dry compared to the moist and wet forests, while the
Fig. 7. (Left) Mean and standard deviation of the reflectance of all pixels within each of the t
between sites.
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)oppositewas true for theNIR and SWIR reflectance (Fig. 6). Broadly spo-
ken we can state that leaf traits related to light capture and growth (for
example, photosynthetic pigments, nutrients and leafmass) are absorb-
ing and scattering light roughly in the 350–700 nm spectral range
(e.g., Asner, 1998; Ollinger, 2011). Secondarymetabolites such as lignin,
cellulose, phenols, and tannins, which contribute to foliar defense and
longevity are active absorbers and scatterers of the NIR and SWIR elec-
tromagnetic energy (e.g., Kokaly, Asner, Ollinger, Martin, & Wessman,
2009; Majeke, van Aardt, & Cho, 2008). This latter spectral region is
also strongly sensitive to water absorption (e.g., Ceccato, Flasse,
Tarantola, Jacquemoud, & Gregoire, 2001).
The observed high variation of VIS reflectance in the dry forests
might as such reflect a strategy to maximize photosynthesis when
nction of kernel size. Note that white areas represent off-the-scale, high CV values.water is available and to minimize water loss and respiration costs dur-
ing rainless periods (Brodribb, Holbrook, Edwards, & Gutierrez, 2003).
Dry forest canopies are indeed characterized by an increasing leaf thick-
ness, decreasing specific leaf area (SLA), shorter leaf life spans, relatively
high P values and more enriched foliar N values suggesting greater re-
sorption and re-metabolism of leaf N in drier forests (Santiago,
Kitajima, & Wright, 2004). Oppositely, it has been reported that spatial
variation in canopy composition in wet forests is strongly driven by
hree sites; (right) pairwise comparison of the spectral similarity (quantified as SI, Eq. 2)
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
Yet, our interpretation explaining spectral variability based on dif-
ferences in leaf traits and tree species richness requires additional con-
siderations on other factors driving spectral variation among the
different forest sites. For example, the relatively high spectral variation
in VIS reflectance observed in the dry forests might to some extent
also be influenced by the observation window. Since the airborne data
were collected during the early dry seasonmany dry forest drought de-
ciduous trees occur in leaf-off conditions exposingmore bark, epiphytes
and dry background to the sensor; components all showing relatively
high VIS reflectance (Clark & Roberts, 2012; Somers, Verbesselt, et al.,
2010; Toomey, Roberts, & Nelson, 2009). In addition, many dry forest
tree species flower in the dry season (Wright & Van Schaik, 1994)
again adding to the VIS reflectance variability (flowers have high VIS re-
flectance; Clark et al., 2005). Another remarkable observation was that
differences in NIR reflectance between the dry site and the wet and
moist sites were relatively small (SI N 15), which could indicate that
on average canopy structure and LAI were comparable. Yet, LiDAR de-
8 B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxpathogens and pests and the higher variation in SWIR reflectance asso-
ciated to a variation in leaf traits related to foliar defense and longevity
is as such not surprising (Asner et al., 2011). Indeed, wet forest canopies
are typically characterized by extended leaf longevity, more structural
defense, higher midday leaf water potential and lower Pmass, Nmass and
SLA (Santiago et al., 2004). These differences in leaf traits for the dry
compared to thewet andmoist site are further highlightedwhen study-
ing the spectral similarity (or species turnover) between the different
forest sites (Fig. 7). Results in Fig. 7 clearly show higher absolute VIS
and SWIR reflectance values (not variability but absolute reflectance
values) for the dry sites.
Along our ecological gradient we also observed an impact on the
spatial patterns in spectral similarity (i.e. beta diversity; Fig. 9). We ob-
served a higher spectral similarity, especially in the VIS and NIR,
throughout the wet compared to the moist and dry forests. The higher
Fig. 8.Average decrease in spectral similarity (SI)with kernel size (i.e. spectral proxy for spe-
cies distance decay curves) and 95% confidence interval for the three different study sites.spectral variation in dry forests could predominantly be linked to habi-
tat fragmentation resulting in increased availability of light resulting in
more pronounced canopy pigmentation and a blend of interior, succes-
sional and invasive species near forest edges (Raghubanshi & Tripathi,
2009; Fig. 10). Also the clear contrast in water content and leaf condi-
tions from riparian drainage to surrounding areas is likely contributing
to the increased spectral variation (Laurance, Ferreira, Rankin-de
Merona, & Laurance, 1998).
WavWavelength (nm)
Fig. 9. Spectral overlap per wavelength (exp
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)rived histograms of the top-of-canopy (TCH) height (Fig. 11) indeed
verified the similar canopy structure between wet (mean TCH =
23.67 m; sd = 7.77 m) and moist sites (mean TCH = 22.13 m; sd =
7.45 m) but revealed a distinctly lower TCH for the dry site (mean
TCH = 10.73 m, sd = 6.37 m). Perhaps high NIR from dry herba-
ceous/soil background and bark exposed in the IFOV elevate NIR in
dry forests to similar levels of moist andwet forests, which in those for-
ests are more likely due to volumetric scattering among leaves. To con-
clude, differences in leaf and reproductive phenology, canopy structure,
and contribution of other components (e.g. epiphylls, bark, back-
ground) are as such additional sources of spectral variation among for-
est sites that contribute to the within-species variability (and spectral
variation; Zhang, Rivard, Sanchez-Azofeifa, & Castro-Essau, 2006) there-
by attenuating the direct link with tree species richness.
This taken into account we can still state that our spectral mesoscale
analysis extends previous results suggesting that niche differentiation
with respect to soil water availability is a direct determinant of both
local- and regional-scale distributions of tropical trees (Condit et al.,
2013). Changes in soil moisture availability caused by global climate
change and forest fragmentation are therefore likely to alter tropical spe-
cies distributions, community composition and diversity (Engelbrecht
et al., 2007; Pyke et al., 2001). We thus contend that the Panamanian
forest shows clear patterns of spatial organization along environmental
gradients, predominantly determined by broad-scale precipitation varia-
tion, but also partly driven bywithin-site variation related to topography
and controls on fine-scale abiotic gradients.
Our results indicate that relative differences in tropical forest canopy
diversity may be monitored using high-resolution imaging spectrosco-
py. A next step would be to test the accuracy and scalability of our re-
sults with lower spatial resolution spectrometer data, simulating the
observing conditions that will be achievedwith future satellitemissions
elength (nm) Wavelength (nm)ressed as SI) as a function of kernel size.
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
0-5 5-10 10-15 15-25
Dry forest site Moist fore
Fig. 10.Mapof the total areaunder the spectral distance decay curve (Fig. 8) for the three study s
9B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxsuch as the European Union's EnMap (Sang et al., 2008) and NASA's
HyspIRImissions. These satellites will observe the land surface at spatial
resolutions of 30–60 m, thereby incorporating multiple tropical forest
canopies into individual measurement pixels. Future research will
therefore focus on performing a comprehensive sensitivity analysis of
spectral diversity measures with respect to spectral mixing and spatial
scaling. These analysis can be performed based on synthetic EnMAP-,
and HyspIRI-like imagery using Carnegie Airborne Observatory as
input of an end-to-end simulation tool like EeteS (Segl et al., 2012).
These synthetic representations at different spatial (and/or spectral
scales) allow for pronouncing the highly amplifiedmixed pixel scenario
typical for coarser resolution spaceborne remote sensing imagery from
tropical areas. Transferring our findings from sub-canopy resolution
spectroscopy from the Carnegie Airborne Observatory to these satellite
missions will require not only developments in remote sensing
methods but also in the way we understand and treat the organisms
(trees, lianas, etc) that comprise the spectroscopic signal at multiple
correlation found in this study, species richness is lower.spatial scales.
Fig. 11. Top-of-canopy height histograms derived from the LiDAR point clouds.
Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)Acknowledgments
The research presented in this paper is funded by the Belgian Science
Policy Office in the framework of the STEREO II Program— Project ReM-
EDy (SR/67/164). The Carnegie Airborne Observatory is made possible
by the Gordon and Betty Moore Foundation, the John D. and Catherine
T. MacArthur Foundation, W. M. Keck Foundation, the Margaret A.
Cargill Foundation, Grantham Foundation for the Protection of the Envi-
ronment, Avatar Alliance Foundation, Mary Anne Nyburg Baker and G.
Leonard Baker Jr., and William R. Hearst III. The plot project is part of
the Center for Tropical Forest Science, a global network of large-scale
demographic tree plots. This research is further supported by VLIR-
UOS (Flemish Interuniversity Council — University Development Coop-
eration) and DGD (the Directorate General for Development Coopera-
tion) (NOPO2014Pr0001) through the KLIMOS consortium.
References
>25
st site Wet forest site
ites. Higher values indicate that spectral similarity is higher and thus, byway of thepositiveAsner, G. P. (1998). Biophysical and biochemical sources of variability in canopy reflec-
tance. Remote Sensing of Environment, 64, 234–253.
Asner, G. P. (2013). Mesoscale exploration and conservation of tropical canopies in a
changing climate. In M. Lowman (Eds.), Treetops at risk: Challenges of global canopy
ecology and conservation. Springer Science + Business Media New York.
Asner, G. P., Anderson, C. B., Martin, R. E., Knapp, D. E., Tupayachi, R., Sinca, F., et al. (2014).
Landscape-scale changes in forest structure and functional traits along an Andes-to-
Amazon elevation gradient. Biogeosciences, 11, 843–856.
Asner, G. P., Knapp, D. E., Boardman, J., Green, R. O., Kennedy-Bowdoin, T., Eastwood, M.,
et al. (2012). Carnegie Airborne Observatory-2: Increasing science data dimensional-
ity via high-fidelitymulti-sensor fusion. Remote Sensing of Environment, 124, 454–465.
Asner, G. P., Knapp, D. E., Kennedy-Bowdoin, T., Jones, M. O., Martin, R. E., Boardman, J.,
et al. (2008). Invasive species detection in Hawaiian rainforests using airborne imag-
ing spectroscopy and LiDAR. Remote Sensing of Environment, 112, 1948–1955.
Asner, G. P., & Martin, R. E. (2009). Airborne spectranomics: Mapping canopy chemical
and taxonomic diversity in tropical forests. Frontiers in Ecology and the Environment,
7, 269–276.
Asner, G. P., Martin, R. E., Knapp, D. E., Tupayachi, R., Anderson, C., Carranza, L., et al.
(2011). Spectroscopy of canopy chemicals in humid tropical forests. Remote Sensing
of Environment, 115, 3587–3598.
Asner, G. P., Nepstad, D., Cardinot, G., & Ray, D. (2004). Drought stress and carbon uptake
in an Amazon forest measured with spaceborne imaging spectroscopy. Proceedings of
the National Academy of Sciences of the United States of America, 101, 6039–6044.
Asner, G. P., & Vitousek, P. M. (2005). Remote analysis of biological invasion and biogeo-
chemical change. Proceedings of the National Academy of Sciences of the United States of
America, 102, 483–486.
Baldeck, C. A., & Asner, G. P. (2013). Estimating vegetation beta diversity from airborne
imaging spectroscopy and unsupervised clustering. Remote Sensing, 5(5), 2057–2071.
Brodribb, T. J., Holbrook, N. M., Edwards, E. J., & Gutierrez, M. V. (2003). Relations between
stomatal closure, leaf turgor and xylem vulnerability in eight tropical dry forest trees.
Plant, Cell & Environment, 26, 443–450.
nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016
Carlson, K. M., Asner, G. P., Hughes, R. F., Ostertag, R., & Martin, R. E. (2007). Hyperspectral Morlon, H., Chuyong, G., Condit, R., Hubbell, S., Kenfack, D., Thomas, D., et al. (2008). A
general framework for the distance-decay of similarity in ecological communities.
Ecology Letters, 11, 904–917.
10 B. Somers et al. / Remote Sensing of Environment xxx (2015) xxx–xxxCarter, G. A., Knapp, A. K., Anderson, J. E., Hoch, G. A., & Smith, M. D. (2005). Indicators of
plant species richness in AVIRIS spectra of a mesic grassland. Remote Sensing of
Environment, 98, 304–316.
Ceccato, P., Flasse, S., Tarantola, S., Jacquemoud, S., & Gregoire, J. M. (2001). Detecting veg-
etation leaf water content using reflectance in the optical domain. Remote Sensing of
Environment, 77, 22–33.
Clark, D. B., & Clark, D. A. (1990). Distribution and effects on tree growth of lianas and
woody hemiepiphytes in a Costa Rican tropical wet forest. Journal of Tropical
Ecology, 6, 321–331.
Clark, M. L., & Roberts, D. A. (2012). Species-level differences in hyperspectral metrics
among tropical rainforest trees as determined by a tree-based classifier. Remote
Sensing, 4, 1820–1855.
Clark, M. L., Roberts, D. A., & Clark, D. B. (2005). Hyperspectral discrimination of tropical rain
forest tree species at leaf to crown scales. Remote Sensing of Environment, 96, 375–398.
Colgan, M., Baldeck, C., Féret, J. B., & Asner, G. P. (2012). Mapping savanna tee species at
ecosystem scales using support vector machine classification and BRDF correction
on airborne hyperspectral and lidar data. Remote Sensing, 4, 3462–3480.
Collwell, R. K., Brehm, G., Cardelus, C. L., Gilman, A. C., & Longino, J. T. (2008). Global
warming, elevational range shifts, and lowland biotic attrition in the wet tropics.
Science, 332, 258–261.
Condit, R. (1998). Tropical forest census plots. Berlin, Germany, and Georgetown, Texas:
Springer-Verlag and R. G. Landes Company.
Condit, R., Ashton, P., Bunyavejchewin, S., Dattaraja, H. S., Davies, S., Esufali, S., et al.
(2006). The importance of demographic niches to tree diversity. Science, 313, 98–101.
Condit, R., Engelbrecht, B. M. J., Pino, D., Pérez, R., & Turner, B. L. (2013). Species distribu-
tions in response to individual soil nutrients and seasonal drought across a commu-
nity of tropical trees. Proceedings of the National Academy of Sciences, 110, 5064–5068.
Condit, R., Pérez, R., & Daguerre, N. (2010). Trees of Panama and Costa Rica. Princeton Uni-
versity Press.
Dietrich, W., Windsor, D., & Dunne, T. (1982). Geology, climate, and hydrology of Barro
Colorado Island. In LeighJr. (Eds.), The ecology of a tropical forest. Washington, DC:
Smithsonian Institution Press.
Duro, D. C., Girard, J., King, D. J., Fahrig, L., Mitchell, S., Lindsay, K., et al. (2014). Predicting
species diversity in agricultural environments using Landsat TM imagery. Remote
Sensing of Environment, 144, 214–225.
Engelbrecht, B. M. J., Comita, L. S., Condit, R., Kursar, T. A., Tyree, M. T., Turner, B. L., et al.
(2007). Drought sensitivity shapes species distribution patterns in tropical forests.
Nature, 447, 80–83.
FAO (2007). State of the world's forests. Rome: FAO.
Feret, J. -B., &Asner, G. P. (2013). Tree species discrimination in tropical forests using airborne
imaging spectroscopy. IEEE Transactions on Geoscience and Remote Sensing, 51, 73–84.
Foster, P. (2001). The potential negative impacts of global climate change on tropical
montane cloud forests. Earth-Science Reviews, 55, 73–106.
Gentry, A. H. (1988). Changes in plant community diversity and floristic composition on en-
vironmental and geographical gradients. Annals of the Missouri Botanical Garden, 75,
1–34.
Gillespie, T.W., Foody, G. M., Rocchini, D., Giorgi, A. P., & Saatchi, S. (2008). Measuring and
modelling biodiversity from space. Progress in Physical Geography, 32, 209–221.
Gotelli, N. J., & Colwell, R. K. (2001). Quantifying biodiversity: Procedures and pitfalls in
the measurement and comparison of species richness. Ecology Letters, 4, 379–391.
Gould, W. (2000). Remote sensing of vegetation, plant species richness, and regional bio-
diversity hotspots. Ecological Applications, 10, 1861–1870.
Higgins,M. A., Asner, G. P.,Martin, R. E., Knapp, D. E., Anderson, C., Kennedy-Bowdoin, T., et al.
(2014). Linking imaging spectroscopy and LiDAR with floristic composition and forest
structure in Panama. Remote Sensing of Environmenthttp://dx.doi.org/10.1016/j.rse.
2013.09.032.
Jusoff, K., & Ibrahim, K. (2009). Hyperspectral remote sensing for tropical rainforest.
American Journal of Applied Sciences, 6, 2001–2005.
Kalacska, M., Sanchez-Azofeifa, G. A., Rivard, B., Caelli, T., White, H. P., & Calvo-Alvarado, J.
C. (2007). Ecological fingerprinting of ecosystem succession: Estimating secondary
tropical dry forest structure and diversity using imaging spectroscopy. Remote
Sensing of Environment, 108, 82–96.
Kokaly, R. F., Asner, G. P., Ollinger, S. V., Martin, M. E., & Wessman, C. A. (2009). Character-
izing canopy biochemistry from imaging spectroscopy and its application to ecosys-
tem studies. Remote Sensing of Environment, 113, S78–S91.
Laurance, W. F., Ferreira, L. V., Rankin-de Merona, J. M., & Laurance, S. G. (1998). Rain for-
est fragmentation and the dynamics of Amazonian tree communities. Ecology, 79,
2032–2040.
Lefsky, M. A., Cohen, W. B., Acker, S. A., Parker, G. G., Spies, T. A., & Harding, D. (1999).
Lidar remote sensing of the canopy structure and biophysical properties of
Douglas-fir western hemlock forests. Remote Sensing of Environment, 70, 339–361.
Levin, N., Shmida, A., Levanoni, O., Tamari, H., & Kark, S. (2007). Predictingmountain plant
richness and rarity from space using satellite-derived vegetation indices. Diversity
and Distributions, 13, 692–703.
Lucas, K., & Carter, G. (2008). The use of hyperspectral remote sensing to assess vascular
plant species richness on Horn Island, Mississippi. Remote Sensing of Environment,
112, 3908–3915.
Majeke, B., van Aardt, J. A. N., & Cho, M. A. (2008). Imaging spectroscopy of foliar bio-
chemistry in forestry environments. Southern Forests, 70, 275–285.Please cite this article as: Somers, B., et al., Mesoscale assessment of cha
Panama using airborne imaging s..., Remote Sensing of Environment (2015)Nagendra, H., & Rocchini, D. (2008). High resolution satellite imagery for tropical biodi-
versity studies: The devil is in the detail. Biodiversity and Conservation, 17, 3431–3442.
Ollinger, S. V. (2011). Sources of variability in canopy reflectance and the convergent
properties of plants. New Phytologist, 189, 375–394.
Palmer, M. W., Earls, P. G., Hoagland, B. W., White, P. S., & Wohlgemuth, T. (2002). Quan-
titative tools for perfecting species lists. Environmetrics, 13, 121–137.
Pau, S., Wolkovich, E. M., Cook, B. I., Nytch, C. J., Regetz, J., Zimmerman, J. K., et al. (2013).
Nature Climate Change, 3, 838–842.
Pyke, C. R., Condit, R., Salamon, A., & Lao, S. (2001). Floristic composition across a climate
gradient in a neotropical lowland forest. Journal of Vegetation Science, 12, 553–566.
R Development Core Team (2010). R: A language and environment for statistical comput-
ing. Available online: http://www.lsw.uni-heidelberg.de/users/christlieb/teaching/
UKStaSS10/R-refman.pdf
Raghubanshi, A. S., & Tripathi, A. (2009). Effect of disturbance, habitat fragmentation and
alien invasive plants on floral diversity in dry tropical forests of Vindhyan highland: A
review. Tropical Ecology, 50, 57–69.
Rand, A., & Rand, W. (1982). Variation in rainfall on Barro Colorado Island. In LeighJr.
(Eds.), The ecology of a tropical forest. Washington, DC: Smithsonian Institution Press.
Rocchini, D. (2007). Effects of spatial and spectral resolution in estimating ecosystem
alpha-diversity by satellite imagery. Remote Sensing of Environment, 111, 423–434.
Rocchini, D., Balkenhol, N., Carter, G. A., Foody, G. M., Gillespie, T. W., He, K., et al. (2010).
Remotely sensed spectral heterogeneity as a proxy of species diversity: Recent ad-
vances and open challenges. Ecological Informatics, 5, 318–329.
Rocchini, D., Butini, S. A., & Chiarucci, A. (2005). Maximizing plant species inventory effi-
ciency by means of remotely sensed spectral distances. Global Ecology and
Biogeography, 14, 431–437.
Rocchini, D., Dadalt, L., Delucchi, L., Neteler, M., & Palmer, M.W. (2014). Disentangling the
role of remotely sensed spectral heterogeneity as a proxy for North American plant
species richness. Community Ecology, 15(1), 37–43.
Rocchini, D., Delucchi, L., Bacaro, G., Cavallini, P., Feilhauer, H., Foody, G. M., et al. (2013).
Calculating landscape diversity with information-theory based indices: A GRASS GIS
solution. Ecological Informatics, 17, 82–93.
Sang, B., Schubert, J., Kaiser, S., Mogulsky, V., Neumann, C., Förster, K. -P., et al. (2008). SPIE
Proceedings, 7086.
Santiago, L. S., Kitajima, K., & Wright, S. J. (2004). Coordinated changes in photosynthesis,
water relations and leaf nutritional traits of canopy trees along a precipitation gradi-
ent in lowland tropical forest. Ecophysiology, 139, 495–502.
Scheiner, S. M. (2003). Six types of species-area curves. Global Ecology and Biogeography,
12, 441–447.
Schimel, D. S., Asner, G. P., & Moorcroft, P. R. (2013). Observing changing ecological diver-
sity in the Anthropocene. Frontiers in Ecology and the Environment, 11, 129–137.
Schnitzer, S. A. (2005). A mechanistic explanation for the global patterns of liana abun-
dance and distribution. The American Naturalist, 166, 262–276.
Segl, K., Guanter, L., Rogass, C., Kuester, T., Roessner, S., Kaufmann, H., et al. (2012).
EeteS — the EnMAP end-to-end simulation tool. IEEE Journal of Selected Topics in
Applied Earth Observations and Remote Sensing, 5, 522–530.
Somers, B., & Asner, G. P. (2012). Hyperspectral time series analysis of native and invasive
species in Hawaiian rainforests. Remote Sensing, 4, 2510–2529.
Somers, B., & Asner, G. P. (2013).Multi-temporalmixture analysis and feature selection for
invasive species mapping in rainforests. Remote Sensing of Environment, 136, 14–27.
Somers, B., Delalieux, S., Stuckens, J., Verstraeten, W. W., & Coppin, P. (2009). A weighted
linear spectral mixture analysis approach to address endmember variability in agri-
cultural production systems. International Journal of Remote Sensing, 30, 139–147.
Somers, B., Delalieux, S., Verstraeten, W. W., van Aardt, J. A. N., Albrigo, G. L., & Coppin, P.
(2010a). An automated wavelength selection technique for optimized hyperspectral
mixture analysis. International Journal of Remote Sensing, 31, 5549–5568.
Somers, B., Verbesselt, J., Ampe, E. M., Sims, N., Verstraeten, W. W., & Coppin, P. (2010b).
Spectral mixture analysis to monitor defoliation in mixed aged Eucalyptus globules
Labill plantations in southern Australia using Landsat 5TM and EO-1 Hyperion data.
International Journal of Applied Earth Observation and Geoinformation, 12, 270–277.
Toomey,M., Roberts, D. A., & Nelson, B. (2009). The influence of epiphylls on remote sens-
ing of humid forests. Remote Sensing of Environment, 113, 1787–1798.
Townsend, A. R., Asner, G. P., & Cleveland, C. C. (2008). The biogeochemical heterogeneity
of tropical forests. Trends in Ecology & Evolution, 23, 424–431.
Ustin, S. L., & Gamon, J. A. (2010). Remote sensing of plant functional types. New
Phytologist, 186, 795–816.
Ustin, S. L., Roberts, D. A., Gamon, J. A., Asner, G. P., & Green, R. O. (2004). Using imaging
spectroscopy to study ecosystem processes and properties. BioScience, 54, 523–534.
Vitousek, P., Asner, G. P., Chadwick, O. A., & Hotchkiss, S. (2009). Landscape-level variation in
forest structure and biogeochemistry across a substrate age in Hawaii. Ecology, 90,
3074–3086.
Wright, S. J. (2005). Tropical forests in a changing environment. Trends in Ecology &
Evolution, 20, 553–560.
Wright, S. J., & Van Schaik, C. P. (1994). Light and the phenology of tropical trees. The
American Naturalist, 143, 192–199.
Zhang, J., Rivard, B., Sanchez-Azofeifa, A., & Castro-Essau, K. (2006). Intra- and inter-class
spectral variability of tropical tree species at La Selva, Costa Rica: Implications for species
identification using HYDICE imagery. Remote Sensing of Environment, 105, 129–141.remote sensing of canopy biodiversity in Hawaiian lowland rainforests. Ecosystems,
10, 536–549.nges in tropical tree species richness across a bioclimatic gradient in
, http://dx.doi.org/10.1016/j.rse.2015.04.016