doi:10.1016/j.rse.2005.08.001Remote Sensing of Environm
Indicators of plant species richness in AVIRIS spectra of a mesic
grassland
Gregory A. Carter a,b,*, Alan K. Knapp c, Jim E. Anderson d, Greg
A. Hoch e, Melinda D. Smith f
aGulf Coast Geospatial Center, The University of Southern
Mississippi, P.O. Box 7000, 703 E. Beach Drive, Ocean Springs, MS
39564, United States bDepartment of Coastal Sciences, The
University of Southern Mississippi, P.O. Box 7000, 703 E. Beach
Drive, Ocean Springs, MS 39564, United States
cDepartment of Biology, Colorado State University, Fort Collins,
CO, United States dApplied Sciences Directorate, NASA, Stennis
Space Center, MS, United States
eDepartment of Biology, Concordia College, Moorhead, MN, United
States fDepartment of Ecology and Evolutionary Biology, Yale
University, New Haven, CT, United States
Received 25 February 2005; received in revised form 3 August 2005;
accepted 6 August 2005
Abstract
Hyperspectral imagery of the Konza Prairie Biological Station in
northeastern Kansas was used to evaluate upwelling spectral
radiance,
prairie spectral reflectance and band ratios of each as potential
indicators of vascular plant species richness in a mesic grassland.
The extent to
which spatial variability in these parameters related to plant
species richness also was investigated. A 224 channel hyperspectral
data cube
acquired in June 2000 by the Airborne Visible and Infrared Imaging
Spectrometer (AVIRIS) provided complete coverage of the
400–2500
nm range at approximately 10 nm per channel. After band deletions
accounted for detector overlap and strong atmospheric
attenuation
features, 176 bands were retained for analysis and spanned the
404–2400 nm range. Prairie reflectance was estimated via radiative
transfer
modeling and scaling to a library spectrum of highway construction
material. Data were sampled from pixels having a 19 m ground
sample
distance (GSD) to represent each of 93 vegetation sampling
transects. Reflectance and radiance at mid-infrared wavelengths
(e.g., 1553 nm),
and band ratios that were based on atmospheric windows in the red,
near-infrared and mid-infrared spectra estimated species richness
to
within 6 to 7 species per transect. The 856 to 780 nm radiance or
reflectance ratio yielded maximum adjusted coefficients of
determination
(r2) of approximately 0.4 in regressions with richness when data
from bison-grazed and ungrazed areas were combined. These
regressions
remained significant ( p0.001) when only ungrazed areas were
assessed although r2 reduced to approximately 0.2. Richness was
related
significantly also to the 433 to 674 nm reflectance ratio for
grazed-plus-ungrazed and ungrazed-only areas. In contrast, the
effectiveness of
the 433 to 674 nm radiance ratio was reduced by atmospheric
backscatter. Species richness did not correlate strongly or
consistently with
transect spatial variability (coefficient of variation or range) in
radiance, reflectance or band ratio value, apparently as a
consequence of the
relatively small area sampled for each transect (approximately 0.5
ha). Relationships between richness and prairie spectral features
were
explained by the influence of soil exposure on both parameters.
Richness and estimated soil exposure tended to increase from
ungrazed
lowlands, to ungrazed slopes, to ungrazed uplands to grazed areas.
Remotely sensed estimates of soil exposure may be particularly
useful in
addressing plant species richness on grazed grasslands owing to an
overall similarity in spectral reflectance among dominant plant
species.
D 2005 Elsevier Inc. All rights reserved.
Keywords: AVIRIS; Biodiversity; Bison grazing; Hyperspectral;
Grassland; Plant species richness; Radiance; Reflectance; Spatial
variability; Spectral mixing;
Soil exposure
0034-4257/$ - see front matter D 2005 Elsevier Inc. All rights
reserved.
doi:10.1016/j.rse.2005.08.001
* Corresponding author. Gulf Coast Geospatial Center, The
University of
Southern Mississippi, P.O. Box 7000, 703 E. Beach Drive, Ocean
Springs,
MS 39564, United States. Fax: +1 228 818 8848.
E-mail address:
[email protected] (G.A. Carter).
1. Introduction
function among organisms, populations and communities, is
increasingly threatened by human activities (Chapin et al.,
2001). The dependence of biodiversity on habitat quality
(Pimm et al., 1995) and climate (Francis & Currie, 2003),
coupled with current rates of habitat loss and increasingly
ent 98 (2005) 304 – 316
strong evidence of anthropogenic climate change (IPCC,
2001), clearly establishes the need for timely assessments of
biodiversity at local to global scales. Such assessments are
necessary to our understanding of biodiversity and the
development of policies and management practices that
foster sustainable ecosystems. Remote sensing has begun to
contribute substantially to this effort, and increasingly
will
facilitate the mapping of species and community distribu-
tions, evaluations of physical constraints on biodiversity,
and
biodiversity monitoring and forecasting (Kerr &
Ostrovsky,
2003; Stoms & Estes, 1993).
detection of organisms or communities or infer their presence
based on habitat characteristics and primary productivity
(Nagendra, 2001; Turner et al., 2003). Habitat or landscape
classifications based on satellite data have been valuable in
estimating mammal (Oindo et al., 2003), bird (Jorgensen &
Nohr, 1996), butterfly (Kerr et al., 2001) and tick (Cumming,
2000) species diversity. Areas of greatest species richness,
the
number of species in a particular location, may coincide for
plant, bird and insect species (Debinski et al., 1999).
Assessing plant biodiversity by remote sensing relies
generally on relationships between species richness and
habitat diversity (Gould, 2000; Nagendra & Gadgil, 1999a)
and has involved a range in measurement scale (Nagendra &
Gadgil, 1999b) and technological approaches. Examples
include the use of environmental and topographic data from
orbital sensors to estimate plant species diversity in an
agricultural landscape (Luoto et al., 2002), and RGB tonal
values from aerial photographs to assess grassland species
diversity (Waldhardt & Otte, 2003). For arctic tundra
(Gould, 2000) and tropical rain forest (Foody & Cutler,
2003) spatial variability in Landsat TM data was linked to
patterns of plant species richness.
Extensive remote sensing research in the central US has
addressed numerous aspects of tallgrass prairie relevant to
patterns and determinants of plant species richness. However,
the evaluation of tallgrass prairie biodiversity by remote
sensing presents a substantial challenge owing in part to the
similarity in spectral reflectance characteristics among the
dominant plant species (Walter-Shea et al., 1992). Spectral
reflectance measured on the ground easily delineated burned
versus unburned prairie (Asrar et al., 1989; Turner et al.,
1992). TM data likewise discriminated among these and other
grassland management practices (Peterson et al., 2002; Price
et al., 2002), and identified cool-versus warm-season cover
types (Guo et al., 2000). Ground spectra (Turner et al.,
1992)
as well as TM data (Guo et al., 2000) indicated responses of
near-infrared reflectance to grazing, the primary influence
on
plant species diversity in tallgrass prairie (Collins et al.,
1998;
Hickman et al., 2004; Risser, 1988). For shortgrass prairie
in
northwestern Kansas, plant species diversity as a function of
grazing intensity was predicted fromTMdata (Lauver, 1997).
Spatial variability in SPOT or TM data as a measure of
landscape texture was sensitive to changes in plant vigor
over
the growing season and correlated with primary production
(Briggs & Nellis, 1991; Knapp et al., 1999). Thus, a
landscape texture approach may hold promise for estimating
plant species richness given apparent relationships of
species
diversity with primary production (Knapp et al., 2002;
Symstad et al., 2003) and landscape heterogeneity (Moser
et al., 2002).
can significantly influence ecosystem productivity (Hooper
& Vitousek, 1997; Smith & Knapp, 2003; Tilman et al.,
1997) with greatest diversity associated with high spatial
heterogeneity of soils or disturbance (Bakker et al., 2003;
Collins et al., 1998; Knapp et al., 1999). Tallgrass prairie
in
particular is characterized by a high plant species diversity
compared with other grasslands (Risser, 1988).
In the present study, we utilized AVIRIS imagery of the
Konza Prairie Biological Station (KPBS), a tallgrass prairie
preserve in northeastern Kansas, to evaluate relationships of
plant species richness with upwelling spectral radiance (L),
prairie spectral reflectance (R) and band ratios in the 404–
2400 nm wavelength range. Denominator band central
wavelengths were selected according to their relatively high
atmospheric transmittance in the visible, near-infrared or
mid-infrared spectral regions. Results for band ratios of L,
which was not corrected for atmospheric interference, were
compared with results for band ratios of R so that the
necessity of correcting for clear-day atmospheric
interference
in this assessment of plant species diversity could be
evaluated. Specific objectives were to: 1) determine the
extent to which species richness correlated directly with L,
R
or band ratio value or their spatial variability; 2) evaluate
the
importance of an approximate correction for atmospheric
interference (scaling to R) in the assessment of diversity,
and
3) interpret significant correlations with respect to prairie
biophysical characteristics and discuss their potential in
remotely sensing plant species diversity on mesic grasslands.
2. Methods
Foundation Long-Term Ecological Research (LTER) site in
northeastern Kansas (39- 5VN, 96- 35VW). All data were
acquired through the KPBS data archives. The KPBS is
subject to a mid-continental climate with average monthly
temperatures ranging from 2.7 -C in January to 26.6 -C in
July. Average annual precipitation is 834 mm with most
occurring during the growing season of April through
September. Although landscape heterogeneity tends to be
relatively low in June (Briggs & Nellis, 1991), earlier
studies
gave no indication that multi-temporal data would improve
the discrimination of tallgrass prairie cover types, which
may
infer differences in species richness, versus the use of mid-
summer data alone (Peterson et al., 2002). Consequently, this
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316306
study was based on the single June 2000 acquisition of
AVIRIS data.
The KPBS is located in the Flint Hills region, which is
characterized by extensive upland and lowland sites and a
range in elevation of approximately 80 m. It is divided into
64
experimental watersheds (e.g., Fig. 1) which are subjected to
differing fire frequency treatments (1, 2, 4, 10 and 20 yr.
burn
intervals; Knapp & Seastedt, 1998). Some fire treatments
have been in effect since 1972 and others began in 1981.
Cattle have been excluded from the KPBS since 1971. In
1987, 30 bison (Bos bison) were reintroduced to 1012 ha
which include 10 watersheds subjected to a range in fire
frequency (1, 2, 4 and 20 yr. intervals). At the time of this
study, bison herd size was approximately 200 individuals.
Vegetation on the KPBS is dominated by the C4 grasses
big bluestem (Andropogon gerardii Vitman), Indiangrass
(Sorghastrum nutans L. Nash), switchgrass (Panicum virga-
tum L.), little bluestem (A. scoparius Michaux), rough
dropseed (Sporobolus asper Michx.) and sideoats grama
(Bouteloua curtipendula Michx.), whereas a variety of C3
grass, forb and woody species constitute most of the plant
species diversity (Freeman, 1998; Smith & Knapp, 2003).
Fire, grazing, climatic variability and topography influence
patterns of plant species richness separately and
interactively.
Richness is lowest on annually burned sites and increases
with decreasing fir frequency (Collins & Steinauer, 1998;
Smith & Knapp, 1999). In contrast, bison grazing
increases
richness, even when fire is frequent (Collins et al., 1998).
Increasing within-season variability in the amount and
frequency of precipitation also has been shown to increase
plant species diversity (Knapp et al., 2002). Across the
topographic gradients at KPBS, species richness is generally
greatest on uplands and least on lowlands that are dominated
by a few C4 grasses (Hartnett & Fay, 1998).
Throughout late May and June, plant species richness,
defined as number of plant species, was recorded for each
Fig. 1. AVIRIS image of the KPBS at a central wavelength of 674
nm.
Highways I-70 (east-to-west, bottom) and K-177 (north-to-south,
right)
are included for reference. White blocks indicate a total of 1181
pixels
from which spectra were extracted to correspond with the locations
of 93
vegetation sampling transects. All transects were located within
the 14
km2 area outlined in black. Watershed boundaries on the KPBS are
shown
in white.
14 km2 area (Fig. 1). The watersheds have been exposed
to a variety of prescribed burning and grazing regimes
since 1982 and 1991, respectively. Ten of the watersheds
were burned annually in spring (six), fall (two) or winter
(two). The remaining two were burned once every four
years. Three of the watersheds were grazed by bison and
the remaining nine were ungrazed. The watersheds include
Florence cherty clay loam soils on uplands, Tully silty clay
loam on lowlands, and slopes with rocky outcrops. Only
transects dominated by grasses and forbs were included in
the study. Transects containing trees and shrubs, which are
found primarily in drainages (Fig. 1) were excluded to
emphasize the predominant grassland community type of
the KPBS. Each transect was comprised of five circular, 10
m2 plots arranged along a 50 m transect starting at the 5 m
point (Bakker et al., 2003). Richness was derived by
compiling data from the five plots and counting the total
number of species sampled. A given species was counted
only once per transect regardless of its frequency within
the transect.
Propulsion Laboratory, occurred at approximately solar
noon on June 22 in a clear sky. Prairie vegetation was
green and vigorous, having received 2.6 and 5.8 cm of
rainfall on June 13 and 20, respectively. Air temperature
ranged from 18.6 -C at pre-dawn to 31.8 -C during mid-
afternoon and averaged 24.9 -C. Surface windspeed during
1100–1600 h CDT ranged from 4.5–6.2 m s1 and
averaged 5.5 m s1. The AVIRIS data cube consisted of
radiance calibrated to 12-bit resolution and stored as 16-bit
integers in raster format within each of 224 spectral bands
that spanned a 374 to 2508 nm range in central wave-
length. Full-width-at-half-maximum (FWHM) bandwidths
were georectified to North American Datum 1983, UTM
Zone 14N (IMAGINE v. 8.3, ERDAS, Inc., Atlanta, GA)
and indicated a 19 m ground sample distance (GSD).
Bands 1–3, 30–32, 96, 106–116, 151–169, and 214–224
were deleted owing to strong atmospheric interference or
detector overlap. This procedure retained 176 bands in the
404 to 2400 nm range for subsequent processing and
analysis. AVIRIS radiance in 16-bit integer format was
divided by channel-dependent gain factors to yield upwell-
ing spectral radiance (L) in units of AW cm2 nm1 sr1.
Beyond the band deletions, no further corrections for
atmospheric absorption or backscatter were applied to the
L spectra.
Institute, Cary, NC) was limited to spectra extracted from
7–19 pixels per transect that represented transect location
and immediately neighboring terrain (0.25 to 0.7 ha). The
total of 1181 spectra for the 93 transects would enable the
regression of species richness with mean transect L, an
approximation of surface reflectance (R), and band ratio
values, and with transect spatial variability (coefficient of
0
10
20
30
40
50
Fig. 2. Measured and simulated spectra used in estimating
prairie
reflectance (R) from AVIRIS radiance (L) and in selecting
denominator
band central wavelengths for ratio computation. (A) Upwelling
radiance
(L) predicted by MODTRAN 4.0 for a 20 km observer altitude
above
ground with nadir view angle, a 100% reflecting ground surface, and
23
km surface visibility (thin curve), and minimum AVIRIS L per band
for
the entire Konza image (thick curve); (B) first-estimate R spectrum
of an I-
70 intersection (thin curve) and R of the same intersection after
linearly
scaling values to match the known spectral R of asphaltic concrete
(Johns
Hopkins University spectral library, ENVI v. 3.6) (thick curve);
(C)
MODTRAN atmospheric transmittance given the input parameters
described for (A). All simulated spectra were resampled to AVIRIS
band
central wavelengths and FWHM.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316 307
variation or range) in these parameters for each spectral
band. The unequal sampling of pixels among transects
resulted from the occasional necessity to avoid nearby
roads, rocky outcrops or woody vegetation. However,
sampling was more uniform among transects than indicated
by the range in number of pixels sampled. On the average,
12.7T2.6 (standard deviation) pixels were sampled per
transect.
atmospheric interference on regressions of species richness
with band ratio values and given that spectroradiometric
data for ground reference targets were not available, prairie
surface reflectance (R) was approximated from L, a
MODTRAN v. 4.0 radiative transfer simulation, dark-pixel
subtraction, and linear scaling to a library reflectance
spectrum of highway construction material. MODTRAN
simulated the upwelling spectral L from a rural landscape
assuming a 100% surface reflectance throughout the 400–
2500 nm wavelength range, a flight altitude of 20 km above
the 400 m elevation of the KPBS, a nadir view angle, and a
ground-level visibility of 23 km. The high-resolution L
spectrum generated by MODTRAN was resampled (ENVI
v. 3.6, Research Systems, Inc., Boulder, CO) to the central
wavelengths and FWHM bandwidths of the AVIRIS bands
(Fig. 2A). A dark-pixel spectrum was extracted from the
KPBS image (minimum pixel value per band, image
statistics, ENVI v. 3.6) to estimate backscatter and noise
from other sources assuming that actual target L approached
zero. In the violet through green and yellow through mid-
infared spectra, these minimum pixel values represented
drainage bottoms and central portions of small ponds,
respectively. A first estimate of R was derived by dividing
AVIRIS L minus dark-pixel L by MODTRAN L minus
dark-pixel L. This quotient was multiplied by 100 to yield R
in percentage units. This procedure yielded a first-estimate
spectral R of an Interstate Highway 70 (I-70) intersection
(Fig. 1, bottom center; Fig. 2B) that was similar to that of
asphaltic concrete (Johns Hopkins University spectral
library, ENVI v. 3.6).
dimension, the I-70 intersection was well-placed as a
reference target at 7- west of nadir. All transects were
located within 1- to 14- west of nadir. Transect-to-reference
distance in the scan-line dimension was a maximum of 2.5
km, and 1.8 km or less for all but 7 transects. Thus, to
further refine the approximation of spectral R, the ground-
level spectral R of the I-70 intersection was assumed to
equal the spectrum of asphaltic concrete. First-estimate R
was multiplied by a linear scaling factor derived for each
band so that the resulting final spectral R of the
intersection
equaled that of asphaltic concrete resampled (ENVI v. 3.6)
to AVIRIS central wavelengths and FWHM bandwidths
(Fig. 2B). This scaling procedure was applied uniformly to
all transect spectra and thus did not account for spatial
variability in atmospheric column water vapor content (Gao
et al., 1993; Qu et al., 2003). However, it yielded a mean R
spectrum that was similar to a summer R spectrum of the
KPBS that was derived earlier from a within-pixel
atmospheric correction of AVIRIS data (Gao et al., 1993)
(Fig. 3). Additionally, the unknown degree of spatial
variability in atmospheric water vapor content did not
appear to be a significant influence on relationships of
species richness with R (see Results and discussion).
All relationships of species richness with L or R were
assessed using simple linear regression analysis (SAS v.
6.12). Band ratios were computed by dividing L or R in
R (
C V
0
10
20
30
40
50
E
B
D
F
Fig. 3. Summary statistics for the 1181 spectra sampled from the
KPBS. (A) Mean upwelling radiance (L) and (B) estimated prairie
reflectance (R), their
standard deviations (C, D) and coefficients of variation (E,
F).
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316308
each band by the same variable in a denominator band that
was selected for its relatively high atmospheric trans-
mittance within the visible (674 nm), near-infrared (780,
875, 1042 or 1240 nm) or mid-infrared spectrum (1553,
1623 or 2141 nm) as indicated by the MODTRAN
simulation described earlier (Fig. 2C). Species richness
was regressed with mean transect L, R, or band ratio value
(n =7–19) for each band. Where a relatively strong
relationship with species richness was found using data
from all 93 transects, the regression procedure was repeated
for each of three randomly-selected sub-samples, each
comprised of approximately 50% of the data (46 transects).
This served to check the consistency with which a particular
L, R, or ratio numerator band yielded a maximal r2 in
regression with richness. Relationships of richness with
spatial variability in L, R, or band ratio value were
evaluated
by regression with transect coefficient of variation (CV) or
data range.
among transects, analytical procedures were repeated using
a reduced data set comprised of only 7 pixel spectra sampled
randomly for each transect from the full data set. Results
based on this uniform sample size were compared with
those for the full data set. This comparison was of
particular
interest in the evaluation of within-transect spatial varia-
bility as an indicator of species richness.
3. Results and discussion
KPBS transect were sampled at microhabitat (0.1 ha) to
within-community (0.1 to 1000 ha) scales, respectively, as
defined previously (Stoms & Estes, 1993). In contrast
with
regional to global scale assessments of biodiversity which
are based on broad heterogeneity among habitats, species
habitat requirements and habitat-type classifications of
remotely-sensed data (Nagendra, 2001; Turner et al.,
2003), our approach utilized relatively fine spatial and
spectral resolutions in comparing species richness directly
with radiance (L) and reflectance (R) spectra within a 14
km2 area characterized by a relatively homogeneous
vegetation type. Although the number of species per transect
ranged from 16 to 61, the vast majority of transects used in
this study were dominated or co-dominated by big bluestem.
Species found on the KPBS in addition to big bluestem and
the other dominants listed under Methods are documented
elsewhere (Freeman, 1998; Towne, 2002). Species observed
in the long-term sampling transects are listed at http://
climate.konza.ksu.edu/konza.
Mean L and R spectra based on all pixels sampled
(n =1181) were similar to AVIRIS spectra reported earlier for
the KPBS in summer (Gao et al., 1993) (Fig. 3). The standard
deviation of this overall mean was generally greatest in
bands
0.0
0.1
0.2
0.3
0.0
0.1
0.2
1583, 0.27
1524, 0.25
1553, 0.24
1553, 0.24
Fig. 4. Adjusted r2 for simple linear regressions of plant species
richness per transect with transect mean R (thick curve) or
transect coefficient of variability
(thin curve) based on a sampling of 7–19 pixels per transect.
Regression analysis was repeated using three random sub-samples,
each comprised of
approximately 50% of the data (46 transects) (A, B, C) and the full
dataset (93 transects) (D). The dotted curves in (D) indicate
adjusted r2 for regressions of
richness with mean R or transect CV which resulted from a uniform
sampling of 7 pixels per transect. Inset numbers indicate central
wavelength and r2,
respectively, at r2 maxima. Where adjusted r20.1, p0.001 if n =93
and p0.018 if n =46. Where adjusted r20.2, p0.0001 if n =93 and
p0.001 if
n =46. In a given band, the same conversion of L to R was applied
to all image pixels. Thus, adjusted r2 for regressions with L were
identical to those shown in
A–D.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316 309
having the greatest L or R. The coefficient of variation, or
standard deviation as a percentage of the mean, was greatest
in the mid-infrared spectrum.
AVIRIS band central wavelengths, adjusted coefficient of
determination (r2), inte
regressions of plant species richness per transect with transect
mean radiance (L)
Central wavelength (nm) L or L-ratio regression
Adjusted r2 a b
Numerator Denominator
Ungrazed only
433 (54) 674 (76) 0.08 61.2 1
856 (83) 780 (80) 0.22 543.8 71
Regression analyses were conducted separately for grazed plus
ungrazed prairie a
Numbers in parentheses are MODTRAN estimates of atmospheric
transmittance
bandwidth (approximately 10 nm FWHM). For all regressions using the
full data
regressions remained significant at p0.0001 although r2 values
decreased (not s
For regressions that were based on data from ungrazed prairie
alone, p0.001.
Simple linear regression indicated significant relation-
ships of plant species richness with mean transect L or R
throughout the mid-infrared spectrum (Fig. 4). Because the
rcept (a), slope (b) and standard error of the estimate (s) for
simple linear
or reflectance (R) at 1553 nm or with selected band ratios
R or R-ratio regression
nd for ungrazed-only prairie.
(percentage) at the central wavelength given the corresponding
AVIRIS
set, the probability of a greater value of the F statistic was
p0.0001. All
hown) when a uniform number of pixels (7) were sampled for all
transects.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316310
conversion of L to R was applied uniformly among spectra,
adjusted r2 for regressions with L were identical to those
shown for R. The occurrence of maximal r2 in the mid-
infrared spectrum was consistent among analyses for three
randomly-selected sub-samples each representing 46 trans-
ects ( p0.018 where adjusted r20.1) and the full data
set representing all 93 transects ( p0.001 where adjusted
r20.1). For the full data set, richness correlated most
strongly and consistently with L or R near 1553 nm where
the adjusted r2=0.24 and standard error of the estimate (s)
was 7 species per transect (Figs. 4 and 6; Table 1). At a
1553 nm central wavelength and FWHM of 11 nm
(AVIRIS band 127) MODTRAN predicted an atmospheric
transmittance of 94% for the conditions described under
Methods (Table 1).
content had been a significant influence on total spatial
variability in L or R in accordance with the spectrally-
dependent influence of water vapor on atmospheric trans-
mittance, r2 for relationships of species richness with L or
R
likewise might have varied with atmospheric transmittance
in spectral regions where water vapor is the predominant
absorber of solar radiation (e.g., 856–1117 and 1454–1783
nm, Gao et al., 1993). A substantial spatial variability in
atmospheric water vapor would tend to increase among-
transect variability in L or R for strongly-absorbed bands
more than for highly-transmitted bands. Consequently, r2
would tend to increase with atmospheric transmittance if
other sources of variability were not predominant. However,
r2 for regressions of richness with R, and thus with L as
described previously, within the 856–1117 and 1454–1783
nm spectra as shown in Fig. 4D were essentially constant
across the broad ranges in MODTRAN-simulated atmos-
pheric transmittance within these regions (Fig. 5). This was
Atmospheric Transmittance (%)
A d ju
856 nm - 1117 nm range
1454 nm - 1783 nm range
Fig. 5. Adjusted r2 for simple linear regressions of plant species
richness
per transect with transect mean R (from Fig. 4D) versus
MODTRAN-
simulated atmospheric transmittance (see Fig. 2C). Adjusted r2 from
only
the 856–1117 and 1454–1783 nm spectra are plotted because
atmospheric
transmittance in these regions is affected predominantly by water
vapor.
The constancy in adjusted r2 value across broad ranges in
atmospheric
transmittance within each spectral region suggests that spatial
variability in
atmospheric water vapor content, which was not accounted for in
scaling to
R, did not influence regression results significantly.
true also for r2 within the 1971–2400 nm range, but these
were not included in Fig. 5 because atmospheric trans-
mittance in this region is influenced substantially by carbon
dioxide and methane as well as water vapor (Gao et al.,
1993). Thus, it appears that spatial variability in
atmospheric
water vapor content was not a significant influence on
present results.
incorporated denominator bands of relatively high atmos-
pheric transmittance for the visible, near-infrared or mid-
infrared spectra yielded r2 and s that were improved only
slightly in most cases compared with results for L or R at
1553 nm (L1553 or R1553) (Fig. 6; Table 1). However,
regressions with R856 /R780, R799 /R875, or the correspond-
ing L ratios yielded greater r2 of approximately 0.4. Range
in ratio value over the full range in richness among all
transects (6116=45 species per transect) was quite small
in some cases (e.g., Fig. 6F). This is explained at least
partially by the 28–76 nm proximity of numerator and
denominator central wavelengths for nearly half of the
ratios listed in Table 1. Nevertheless, band central wave-
lengths that were optimal for the numerator were similar
among results for sub-samples and the full data set
whether the ratio value range was relatively small or large
(e.g., Fig. 7). For this analysis of the full data set, r2 and
s
for L-ratio regressions tended to be similar to those from
R-ratio regressions when only near- and mid-infrared
bands were used (Fig. 7H; Table 1). However, L-ratios
that incorporated shorter-wavelength bands in the visible
spectrum produced different regression results compared
with backscatter-corrected R ratios (Fig. 7G; Table 1).
Thus, although the determination of R was approximate,
results emphasize the importance of correction for visible-
spectrum backscatter in interpreting spectral indicators of
species richness on the KPBS.
In contrast to results based on mean transect L, R or band
ratio value, species richness did not correspond strongly or
consistently with spatial CV or data range (Figs. 4 and 6;
Table 2). Although regressions involving the CV or range
for some band ratios were significant ( p =0.05), adjusted r2
were generally very low at 0.1 or less. However, this
approach addressed spatial heterogeneity in prairie spectral
features for relatively small areas on the order of 0.5 ha.
Apparently, these areas were not sufficiently large to
capture
a landscape heterogeneity, or texture, that might indicate
trends in species richness. Earlier studies on the KPBS
addressed larger areas that ranged from approximately 10 ha
to several thousand ha and reported a sensitivity of textural
indices to seasonal changes in plant vigor and primary
productivity (Briggs & Nellis, 1991; Knapp et al., 1999).
Given the inverse relationship of aboveground biomass
production in prairie vegetation with plant species diversity
(Knapp et al., 2002), such measures of landscape hetero-
geneity might also indicate species richness. In the present
case, including the CV or range of a NDVI computed from
the AVIRIS data as a surrogate for biomass together with
L 856
/L 780
L 433
/L 674
P la
s p e c ie
s p
e r
10
30
50
70
-1 sr -1)
1.18 1.38 1.58
10
30
50
70
2 = 0.11
Fig. 6. Simple linear regressions of plant species richness per
transect with transect mean L or R at 1553 nm (A, B), transect mean
of the indicated band ratio of
L or R (C–F), and the transect coefficient of variation (CV) for
the indicated band ratios (G, H). Probabilities of a greater value
of the F statistic were
p0.0001 for A–F and p0.02 for G and H. Data from bison-grazed and
ungrazed prairie are represented by dark circles and open
triangles, respectively.
These particular regressions are shown because (A) and (B) produced
the greatest r2 based on L or R per se, (C–F) can be compared with
Fig. 8 and (E) and (F)
produced the greatest r2 overall, and (G) and (H) produced the
greatest r2 based on within-transect spatial variability.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316 311
A dj
us te
d r
E
B
D
F
HG
856, 0.49
866, 0.49
856, 0.38
452, 0.36
442, 0.30
433, 0.29
433, 0.33
856, 0.40
646, 0.28
646, 0.22
646, 0.37
646, 0.33
Fig. 7. Adjusted r 2 for simple linear regressions of plant species
richness per transect with transect mean band ratios of R. Ratios
were computed by dividing R
in each band by R at 674 nm (A, C, E, G) or 780 nm (B, D, F, H).
Inset numbers indicate numerator band central wavelength and r2,
respectively, at r2 maxima.
Regressions were repeated using three sub-samples, each comprised
of approximately 50% of the data (46 transects) (A–F) and the full
dataset (93 transects)
(G, H). Where adjusted r20.1, p0.001 if n =93 and p0.018 if n =46.
Where adjusted r20.2, p0.0001 if n =93 and p0.001 if n =46. Dotted
curves
in (G) and (H) indicate results based on L ratios rather than R
ratios. These results exemplify the approximate consistency of
numerator central wavelength at r2
maximum among regressions that were based on sub-samples and the
full data set. Results for denominator wavelengths of 674 and 780
nm are presented to
show this consistency for ratios characterized by relatively large
versus small numerical ranges in ratio value, respectively (Fig.
6C–F).
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316312
CV or range of the ratio values described earlier in
regressions with richness yielded no significant improve-
ment in r2 (results not shown). This also may have been due
to the small areas sampled, but relationships of the NDVI
with biomass on the KPBS can be site-specific and differ
between years (Weiser et al., 1986) or seasonally (Turner et
al., 1992).
Regressions that were significant when the full data set
was used tended to remain so when a uniform 7 pixels were
sampled for each transect, although p values increased and
adjusted r2 decreased (Fig. 4; Tables 1 and 2 legends).
Transect means of L and R computed from a greater
sampling of pixels apparently were more representative of
prairie features that corresponded with richness.
The range in species richness encountered was influ-
enced substantially by including grazed along with ungrazed
prairie in the analysis (Fig. 6). Consequently, analytical
procedures were repeated separately for grazed versus
ungrazed areas. For grazed areas, regressions of richness
with mean transect L, R, or band ratio value, or with
transect
Table 2
AVIRIS band central wavelengths, adjusted coefficient of
determination (r2), intercept (a), slope (b) and standard error of
the estimate (s) for simple linear
regressions of plant species richness per transect with transect
spatial variability (CV or range) in radiance (L) or reflectance
(R) band ratio values
Central wavelength (nm) L ratio variation regression R ratio
variation regression
Numerator Denominator Adjusted r2 a b s Adjusted r2 a b s
Within-transect CV
1693 (92) 1623 (94) 0.09 49.0 19.2 7.9 0.10 49.7 16.9 7.8
2091 (89) 2141 (96) 0.07 44.5 4.2 8.0 0.11 45.4 3.6 7.8
Within-transect range
1693 (92) 1623 (94) 0.07 46.7 555.6 8.0 0.09 47.6 400.2 7.9
2091 (89) 2141 (96) 0.05 43.8 138.5 8.1 0.08 88.5 112.1 8.0
Results were based on data for grazed plus ungrazed prairie.
Numbers in parentheses are MODTRAN estimates of atmospheric
transmittance (percentage) at the central wavelength given the
corresponding AVIRIS
bandwidth (approximately 10 nm FWHM). For all regressions, the
probability of a greater value of the F statistic was p0.02. All
regressions remained
significant at p0.05 although r2 values decreased (not shown) when
a uniform number of pixels (7) were sampled for all
transects.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316 313
CVor range in L, R, or ratio value, produced adjusted r2 that
remained consistently below 0.2 (results not shown). For
ungrazed areas, this was true also with the exceptions that
r2
were 0.20 or greater for regressions of richness with R433 /
R674, R856 /R780 and L856 /L780 (Fig. 8; Table 1).
Apparently,
interference from atmospheric backscatter resulted in a
much lower adjusted r2 for regression with L433 /L674 (Fig.
8A). Within the ungrazed areas, an influence of topography
on richness and ratio value was observed (Fig. 8).
Consequently, results for grazed plus ungrazed areas and
for ungrazed-only areas may be explained largely by
L 856
/L 780
P la
s p e c ie
s p
e r
15
35
r 2 = 0.08
r 2 = 0.21
Fig. 8. Linear regressions of plant species richness per transect
with the indicate
Probabilities of a greater value of the F statistic were p0.001.
Symbols represe
uplands (closed triangles).
features.
and increases its spatial variability owing to a
characteristic
patchiness in grazing intensity and consequent patchiness in
fuel availability and fire intensity (Knapp et al., 1999).
Indeed, lower R in the near-infrared and increased mid-
infrared R for grazed versus ungrazed prairie (Fig. 9A) is
consistent with an increase in bare soil or senescent
vegetation on the KPBS (Asrar et al., 1986). Mean spectral
R of ungrazed slopes, ungrazed uplands and all grazed areas
R 856
/R 780
r 2 = 0.25
r 2 = 0.20
d transect mean L or R band ratio for ungrazed prairie (see Fig.
5C–F).
nt transects located on lowlands (closed circles), slopes (open
squares) and
Wavelength (nm)
B
Fig. 9. Mean spectral reflectance (R) of ungrazed lowland prairie
(thin solid
curve with maximum near-infrared R), ungrazed slopes (thick solid
curve)
and all bison-grazed areas regardless of topography (dotted curve)
derived
from AVIRIS data, along with a mean R for dried laboratory samples
of soil
types found commonly on the KPBS (thinnest curve, minimum
near-
infrared R) (A). Differences in R (B) determined by subtracting the
mean R
of ungrazed slopes, all grazed areas or ungrazed uplands from a
linear
spectral mixture comprised of 87% ungrazed lowland R plus 13% soil
R
(thick solid curve), 81% ungrazed lowland R plus 19% soil R (dotted
curve)
or 73% ungrazed lowland R plus 27% soil R (thin solid curve),
respectively.
Soil spectra were obtained by the FIFE Information System staff
from
Stoner et al. (1980) and provided by the Oak Ridge National
Laboratory,
Distributed Active Archive Center (www.daac.ornl.gov). Owing to
short-
wavelength noise in the original spectra, mean soil R at
wavelengths of 490
nm or less was assumed to be 3%.
R856/R780
R433/R674
15
25
35
45
S im
u la
a re
( %
)
20
25
30
A
B
Fig. 10. Simulated relationships of soil exposure with R433 /R674
and R856 /
R780. These estimated trends (solid lines) were based on linear
spectral
mixtures of mean R for ungrazed lowland prairie (lowest soil
exposure)
with a mean soil R derived from data of Stoner et al. (1980) (Fig.
9). Mean
values of R433 /R674 and R856 /R780 for ungrazed slopes (open
squares),
ungrazed uplands (closed triangles) and all grazed areas (open
triangles) are
plotted on the trend lines (compare with Figs. 6C–F and 8).
Plant Species Richness (species per 50 m2)
20 30 40 50 60
0 0.5 km
Fig. 11. Plant species richness predicted from its relationship
with R856 /
R780 for a 14 km 2 area on the Konza Prairie Biological Station.
The location
of this area is shown in Fig. 1. White lines indicate watershed
boundaries.
G.A. Carter et al. / Remote Sensing of Environment 98 (2005)
304–316314
could be simulated to within 4 percentage units or less by
linear spectral mixtures of mean ungrazed lowland R and a
mean soil R spectrum that was computed from laboratory
spectra of sieved samples representing KPBS soil types
(Fig. 9B). When mean values of R433 /R674 and R856 /R780
for ungrazed slopes, ungrazed uplands and all grazed areas
were plotted on linear responses of ratio value to soil
exposure that were simulated by spectral mixing, soil
exposure was estimated to have increased from ungrazed
lowlands (unknown but minimum soil exposure) to
ungrazed slopes, to ungrazed uplands, to grazed areas
(Fig. 10). This indicated that soil exposure was 18–21%
greater on ungrazed slopes, 25% greater on ungrazed
uplands, and 26–45% greater on grazed areas than on
ungrazed lowlands. With respect to grazed areas, this
estimate of soil exposure is consistent with the target
consumption of aboveground biomass by bison of 25–40%
(Knapp et al., 1999). Additionally, when the mixing
procedure was applied to simulate responses of the variables
in Table 1 to soil exposure, simulated slope inclinations
(positive versus negative) were identical to those observed.
Consequently, it appears that the observed correlations of
plant species richness with prairie spectral characteristics
can be explained by degree of soil exposure and the known
linkage between richness and disturbance (Knapp et al.,
1999). Based on its relationship with R856 /R780, richness
was predicted for the 14 km2 area that contained all 93
transects (Fig. 11).
Plant species richness on the KPBS could be estimated to
precisions of 6 to 7 species within a range of 16 to 61
species
per transect by L or R at mid-infrared wavelengths (e.g.,
1553 nm) or by band ratios that were based on atmospheric
windows in the red, near-infrared and mid-infrared spectra.
Adjusted r2 were virtually identical for regressions of
richness with L versus R band ratios when highly-trans-
mitted near- or mid-infrared bands were used in the
denominator (Table 1). Thus, the additional data processing
required to derive R would not have been necessary for
identifying the band ratios that best correlated with
richness.
This suggests that ratios of highly-transmitted L bands might
be used in rapid assessments of prairie biodiversity,
avoiding
the additional expense of processing to R.
The ratios L856 /L780 and R856 /R780 yielded the greatest
adjusted r2 of approximately 0.4 when data from grazed
versus ungrazed areas were combined. Regressions of
richness with these band ratios remained highly significant
( p0.001) although r2 decreased to approximately 0.2
when only ungrazed areas were considered. Richness also
regressed significantly with R433 /R674 for grazed-plus-
ungrazed and ungrazed-only areas. By comparison, the
correlation of L433 /L674 with richness was reduced substan-
tially, indicating the necessity of corrections for
atmospheric
backscatter in the short-wavelength visible spectrum.
Species richness was not related strongly or consistently
with transect spatial variability (CV or range) in L, R, or
band ratio value. This contrasts with results of earlier
studies
that addressed spatial variability in larger geographic areas
on the KPBS, but likely was due to the relatively small area
sampled for each transect (approximately 0.5 ha).
Relationships of species richness with L, R or band
ratios were explained by the apparent influence of soil
exposure on plant species richness and prairie spectral
properties. Richness and estimated degree of soil exposure
tended to increase from ungrazed lowlands, to ungrazed
slopes, to ungrazed uplands to grazed areas. The use of
AVIRIS data and scaling to R proved advantageous in
simulating changes in spectral R associated with top-
ography and estimating degree of soil exposure via spectral
mixtures of ungrazed lowland R and a library spectrum of
soil R. Given the consistency with which dry soil
reflectance increases with wavelength, this might have
been done as well using broader spectral bands. However,
the hyperspectral data also enabled regressions of richness
with narrow-band ratios that incorporated proximal numer-
ator and denominator central wavelengths and indicated
richness most effectively, such as L856 /L780 and R856 /R780.
Overall, the remote sensing of soil exposure may be
particularly useful as an indicator of species richness in
grazed grasslands owing to an overall similarity in spectral
reflectance among dominant plant species.
Acknowledgments
vegetation data were provided by the NSF Long Term
Ecological Research Program at Konza Prairie Biological
Station. Soil reflectance spectra were extracted by FIFE
Information System staff from Stoner et al. (1980) and
obtained through the Oak Ridge National Laboratory
Distributed Active Archive Center (www.daac.ornl.gov).
The permission of Stoner et al. (1980) to use these data is
greatly appreciated. The authors thank Alan Criss, Marc
Foster, Debra Armstead, Chris Brown, Paul Kay, Les
Graham, Cathy Schloss and three anonymous reviewers for
their valuable assistance.
References
Asrar, G., Myneni, R. B., Li, Y., & Kanemasu, E. T. (1989).
Measuring and
modeling spectral characteristics of a tallgrass prairie. Remote
Sensing
of Environment, 27, 143–155.
Asrar, G., Weiser, R. L., Johnson, D. E., Kanemasu, E. T., &
Killeen, J. M.
(1986). Distinguishing among tallgrass prairie cover types
from
measurements from multispectral reflectance. Remote Sensing
of
Environment, 19, 159–169.
Bakker, C., Blair, J. M., & Knapp, A. K. (2003). Does resource
availability,
resource heterogeneity or species turnover mediate changes in
plant
species richness in grazed grasslands. Oecologia, 137,
385–391.
Briggs, J. M., & Nellis, M. D. (1991). Seasonal variations of
heterogeneity
in the tallgrass prairie: A quantitative measure using remote
sensing.
Photogrammetric Engineering and Remote Sensing, 57, 407–411.
Chapin III, F. S., Sala, O., Huber-Sannwald, E., & Leemans, R.
(2001). The
future of biodiversity in a changing world. In F. S. Chapin, O.
Sala, & E.
Huber-Sannwald (Eds.), Global biodiversity in a changing
environ-
ment: Scenarios for the 21st century (pp. 1–4). New York’
Springer-
Verlag.
Collins, S. L., Knapp, A. K., Briggs, J. M., Blair, J. M., &
Steinauer, E. M.
(1998). Modulation of diversity by grazing and mowing in
native
tallgrass prairie. Science, 280, 745–747.
Collins, S. L., & Steinauer, E. M. (1998). Disturbance,
diversity, and
species interactions in tallgrass prairie. In A. K. Knapp, J. M.
Briggs, D.
C. Hartnett, & S. L. Collins (Eds.), Grassland dynamics:
Long-term
ecological research in tallgrass prairie (pp. 140–156). New
York’
Oxford University Press.
Cumming, G. S. (2000). Using habitat models to map diversity:
Pan-
African species richness of ticks (Acari: Ixodida). Journal of
Biogeog-
raphy, 27, 425–440.
Debinski, K. D., Kindscher, M., & Jakubauskas, M. E. (1999). A
remote
sensing and GIS-based model of habitats and biodiversity in the
Greater
Yellowstone Ecosystem. International Journal of Remote Sensing,
17,
3281–3291.
Foody, G. M., & Cutler, M. E. J. (2003). Tree biodiversity in
protected and
logged Bornean tropical rain forests and its measurement by
satellite
remote sensing. Journal of Biogeography, 30, 1053–1066.
Francis, A. P., & Currie, D. J. (2003). A globally consistent
richness-
climate relationship for angiosperms. The American Naturalist,
161,
523–536.
Freeman, C. C. (1998). The flora of the Konza Prairie: A historical
review
and contemporary patterns. In A. K. Knapp, J. M. Briggs, D.
C.
Hartnett, & S. L. Collins (Eds.), Grassland dynamics:
Long-term
ecological research in tallgrass prairie (pp. 69–80). New York’
Oxford
University Press.
Gao, B. -C., Heidebrecht, K. B., & Goetz, A. F. H. (1993).
Derivation of
scaled surface reflectances from AVIRIS data. Remote Sensing
of
Environment, 44, 165–178.
Gould, W. (2000). Remote sensing of vegetation, plant species
richness,
and regional biodiversity hotspots. Ecological Applications,
10,
1861–1870.
Guo, X., Price, K. P., & Stiles, J. M. (2000). Biophysical and
spectral
characteristics of cool- and warm-season grasslands under three
land
management practices in eastern Kansas. Natural Resources
Research,
9, 321–331.
Hartnett, D. C., & Fay, P. A. (1998). Plant populations:
Patterns and
processes. In A. K. Knapp, J. M. Briggs, D. C. Hartnett, & S.
L. Collins
(Eds.), Grassland dynamics: Long-term ecological research in
tallgrass
prairie (pp. 81–100). New York’ Oxford University Press.
Hickman, K. R., Hartnett, D. C., Cochran, R. C., & Owensby, C.
E. (2004).
Grazing management effects on plant species diversity in
tallgrass
prairie. Journal of Range Management, 57, 58–65.
Hooper, D. U., & Vitousek, P. M. (1997). The effects of plant
composition
and diversity on ecosystem processes. Science, 277,
1302–1305.
Intergovernmental Panel on Climate Change. (2001). Climate
change,
synthesis report, third assessment report of the intergovernmental
panel
on climate change. New York’ Cambridge University Press.
Jorgensen, A. F., & Nohr, H. (1996). The use of satellite
images for
mapping of landscape and biological diversity in the Sahel.
Interna-
tional Journal of Remote Sensing, 17, 91–109.
Kerr, J. T., & Ostrovsky, M. (2003). From space to species:
Ecological
applications for remote sensing. Trends in Ecology and Evolution,
18,
299–305.
Kerr, J. T., Southwood, T. R. E., & Cihlar, J. (2001). Remotely
sensed
habitat diversity predicts butterfly species richness and
community
similarity in Canada. Proceedings of the National Academy of
Sciences,
98, 11365–11370.
Knapp, A. K., Blair, J. M., Briggs, J. M., Collins, S. L.,
Hartnett, D. C.,
Johnson, L. C., et al. (1999). The keystone role of bison in
north
American tallgrass prairie. Bioscience, 49, 39–50.
Knapp, A. K., Fay, P. A., Blair, J. M., Collins, S. L., Smith, M.
D.,
Carlisle, J. D., et al. (2002). Rainfall variability, carbon
cycling,
and plant species diversity in a mesic grassland. Science,
298,
2202–2205.
Knapp, A. K., & Seastedt, T. R. (1998). Grasslands, Konza
Prairie and
long-term ecological research. In A. K. Knapp, J. M. Briggs, D.
C.
Hartnett, & S. L. Collins (Eds.), Grassland dynamics:
Long-term
ecological research in tallgrass prairie (pp. 3–15). New York’
Oxford
University Press.
Lauver, C. L. (1997). Mapping species diversity patterns in the
Kansas
shortgrass region by integrating remote sensing and vegetation
analysis.
Journal of Vegetation Science, 8, 387–394.
Luoto, M., Toivonen, T., & Heikkinen, R. K. (2002). Prediction
of total and
rare plant species richness in agricultural landscapes from
satellite
images and topographic data. Landscape Ecology, 17, 195–217.
Moser, D., Zechmeister, H. G., Plutzar, C., Sauberer, N., Wrbka,
T., &
Grabherr, G. (2002). Landscape patch shape complexity as an
effective
measure for plant species richness in rural landscapes.
Landscape
Ecology, 17, 657–669.
Nagendra, H. (2001). Using remote sensing to assess biodiversity.
Interna-
tional Journal of Remote Sensing, 22, 2377–2400.
Nagendra, H., & Gadgil, M. (1999a). Satellite imagery as a tool
for
monitoring species diversity: An assessment. Journal of
Applied
Ecology, 36, 388–397.
Nagendra, H., & Gadgil, M. (1999b). Biodiversity assessment at
multiple
scales: Linking remotely sensed data with field information.
Proceed-
ings of the National Academy of Sciences USA, 96, 9154–9158.
Oindo, B. O., Skidmore, A. K., & DeSalvo, P. (2003). Mapping
habitat and
biological diversity in the Maasai Mara ecosystem.
International
Journal of Remote Sensing, 24, 1053–1069.
Peterson, D. L., Price, K. P., & Martinko, E. A. (2002).
Discriminating
between cool season and warm season grassland cover types in
northeastern Kansas. International Journal of Remote Sensing,
23,
5015–5030.
Pimm, S. L., Russell, G. J., Gittleman, J. L., & Brooks, T. M.
(1995). The
future of biodiversity. Science, 269, 347–350.
Price, K. P., Guo, X., & Stiles, J. M. (2002). Optimal Landsat
TM band
combinations and vegetation indices for discrimination of six
grassland
types in eastern Kansas. International Journal of Remote Sensing,
23,
5031–5042.
Qu, Z., Kindel, B. C., & Goetz, A. F. H. (2003). The high
accuracy
atmospheric correction for hyperspectral data (HATCH) model.
IEEE
Transactions on Geoscience and Remote Sensing, 41, 1223–1231.
Risser, P. G. (1988). Diversity in and among grasslands. In E. O.
Wilson
(Ed.), Biodiversity (pp. 176–180). Washington D.C.’ National Acad-
emy Press.
Smith, M. D., & Knapp, A. K. (1999). Exotic plant species in a
C-4-
dominated grassland: Invasibility, disturbance, and community
struc-
ture. Oecologia, 120, 605–612.
Smith, M. D., & Knapp, A. K. (2003). Dominant species
maintain
ecosystem function with non-random species loss. Ecology Letters,
6,
509–517.
Stoms, D. M., & Estes, J. E. (1993). A remote sensing research
agenda for
mapping and monitoring biodiversity. International Journal of
Remote
Sensing, 10, 1839–1860.
Stoner, E. R., Baumgardner, M. F., Biehl, L. L., & Robinson, B.
F. (1980).
Atlas of soil reflectance properties. Agricultural Research
Station
Research Bulletin, vol. 962. West Lafayette, IN’ Purdue
University.
Symstad, A. J., Chapin III, F. S., Wall, D. H., Gross, K. L.,
Huenneke, L. F.,
Mittelbach, G. G., et al. (2003). Long-term and large-scale
perspectives
on the relationship between biodiversity and ecosystem
functioning.
Bioscience, 53, 89–98.
Tilman, D., Knops, J., Wedin, D., Reich, P., Ritchie, M., &
Sieman, E.
(1997). The influence of functional diversity and composition
on
ecosystem processes. Science, 277, 1300–1302.
Towne, E. G. (2002). Vascular plants of Konza Prairie Biological
Station:
An annotated checklist of species in a Kansas tallgrass prairie.
Sida, 20,
269–294.
Turner, C. L., Seastedt, T. R., Dyer, M. I., Kittel, T. G. F.,
& Schimel, D. S.
(1992). Effects of management and topography on the
radiometric
response of a tallgrass prairie. Journal of Geophysical Research,
97,
18855–18866.
Turner, W., Spector, S., Gardiner, N., Fladeland, M., Sterling, E.,
&
Steininger, M. (2003). Remote sensing for biodiversity science
and
conservation. Trends in Ecology and Evolution, 18, 306–314.
Waldhardt, R., & Otte, A. (2003). Indicators of plant species
and
community diversity in grasslands. Agriculture, Ecosystems
and
Environment, 98, 339–351.
Walter-Shea, E. A., Blad, B. L., Hays, C. J., Mesarch, M. A.,
Deering, D.
W., & Middleton, E. M. (1992). Biophysical properties
affecting
vegetative canopy reflectance and absorbed photosynthetically
active
radiation at the FIFE site. Journal of Geophysical Research,
97,
18925–18934.
Weiser, R. L., Asrar, G., Miller, G. P., & Kanemasu, E. T.
(1986). Assessing
grassland biophysical characteristics from spectral
measurements.
Remote Sensing of Environment, 20, 141–152.
Indicators of plant species richness in AVIRIS spectra of a mesic
grassland
Introduction
Methods