IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 39, NO. 7, JULY 2001 1491
Scaling-Up and Model Inversion Methods with
Narrowband Optical Indices for Chlorophyll
Content Estimation in Closed Forest Canopies with
Hyperspectral Data
Pablo J. Zarco-Tejada, John R. Miller, Thomas L. Noland, Gina H. Mohammed, and Paul H. Sampson
Abstract—Radiative transfer theory and modeling assumptions
were applied at laboratory and field scales in order to study the
link between leaf reflectance and transmittance and canopy hyper-
spectral data for chlorophyll content estimation. This study was
focused on 12 sites of Acer saccharum M. (sugar maple) in the Al-
goma Region, Canada, where field measurements, laboratory-sim-
ulation experiments, and hyperspectral compact airborne spectro-
graphic imager (CASI) imagery of 72 channels in the visible and
near-infrared region and up to 1-m spatial resolution data were
acquired in the 1997, 1998, and 1999 campaigns. A different set
of 14 sites of the same species were used in 2000 for validation of
methodologies. Infinite reflectance and canopy reflectance models
were used to link leaf to canopy levels through radiative transfer
simulation. The closed and dense (
4
) forest canopies of
Acer saccharum M. used for this study, and the high spatial resolu-
tion reflectance data targeting crowns, allowed the use of optically
thick simulationformulaeandturbid-medium SAILHand MCRM
canopy reflectance models for chlorophyll content estimation by
scaling-up and by numerical model inversion approaches through
coupling to the PROSPECT leaf radiative transfer model. Study of
the merit function in the numerical inversion showed that red edge
optical indices used inthe minimizingfunction suchas
750 710
perform better than when all single spectral reflectance channels
from hyperspectral airborne CASI data are used, and in addition,
the effect of shadows and LAI variation are minimized. Estimates
of leaf pigment by hyperspectral remote sensing of closed forest
canopies were shown to be feasible with root mean square errors
(RMSE’s) ranging from 3 to
5 5
g cm
2
. Pigment estimation by
model inversion as described in this paper using these red edge in-
Manuscript received September 2, 2000; revised February 6, 2001. This work
was supported in part bytheCentre for Research in Earth and Space Technology
(CRESTech), the Ontario Ministry of Natural Resources, the Canadian Forestry
Service, the Ministry of Environment and Energy, and Geomatics for Informed
Decisions (GEOIDE), part of the Canadian Networks of Centres of Excellence
Programme.
P. J. Zarco-Tejada was with the Centre for Research in Earth and Space Sci-
ence (CRESS), York University, Toronto, ON M3J 1P3, Canada. He is now with
the Department of Land, Air, and Water Resources (LAWR) Center for Spatial
Technologies and Remote Sensing (CSTARS), University of California, Davis,
CA 95616-8671 USA (e-mail: pzarco@ucdavis.edu).
J. R. Miller is with the Centre for Research in Earth and Space Science
(CRESS), York University, Toronto, ON M3J 1P3, Canada, and also with the
Department of Physics and Astronomy, York University, Toronto, ON M3J
1P3, Canada (e-mail: jrmiller@yorku.ca).
T. L. Noland is with the Ontario Forest Research Institute, Ontario Ministry
of Natural Resources, Sault Ste. Marie, ON P6A 2E5, Canada.
G. H. Mohammed is with the Ontario Forest Research Institute, Ontario Min-
istry of Natural Resources, Sault Ste. Marie, ON P6A 2E5, Canada, and also
with P&M Technologies, Sault Ste. Marie, ON P6A 6S7, Canada.
P. H. Sampson is with the Ontario Forest Research Institute, Ontario Ministry
of Natural Resources, Sault Ste. Marie, ON P6A 2E5, Canada.
Publisher Item Identifier S 0196-2892(01)05497-3.
dices can in principle be readily transferred to the MERIS sensor
using the
750 705
optical index.
Index Terms—Chlorophyll, hyperspectral, leaf reflectance, op-
tical indices, radiative transfer.
I. INTRODUCTION
E
XTENSIVE research has been carried out at the leaf level
in order to assess the physiological condition based on the
study of the light interaction with the foliar medium. The total
chlorophyll content in leaves decreases in stressed vegetation,
changing the proportion of light-absorbing pigments and
leading to less overall absorption with chlorophyll
and
( ) being the most important plant pigments absorbing
blue and red light in the 430–660 nm region, respectively [1],
[2]. Differences in reflectance between healthy and stressed
vegetation due to changes in pigment levels have been detected
in the green peak and along the red edge (690 to 750 nm) (e.g.,
[3]–[6]), allowing remote detection methods to identify vege-
tation stress and mapping through the influence of chlorophyll
content variation [7]. Several narrowband leaf-level optical
indices have been reported in the literature that might be applied
to hyperspectral canopy reflectance data for
estimation
at larger scales [8], [9]. Nevertheless, most studies related
to optical indices for vegetation functioning are based on
measurements made at the leaf level rather than at the canopy
level, where correlation between chlorophyll content and
spectral reflectance can be readily observed [10]–[14]. These
potentially valuable optical indices, both traditional and new
developed narrowband indices derived from recent research
based on reflectance and derivative spectra, are grouped into
four categories, based on the spectral region and the type of
parameter used [8], [9], [15]–[17].
1) Visible Ratios: SRPI
; NPQI
; PRI calculated as
,
and ; NPCI
; Carter ,G
and Lichtenthaler ;
2) Visible/NIR Ratios:NDVI
;
SR
; Lichtenthaler
, ; and SIPI
;
0196–2892/01$10.00 ©2001 IEEE
1492 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 39, NO. 7, JULY 2001
3) Red Edge Reflectance-Ratio Indices: Vogelmann
, ,
; Gitelson and Merzylak
, ; Carter ; cur-
vature index
, and the area of the
derivative under the red edge
;
4) Spectral and Derivative Red Edge Indices: the red
edge inflection and chlorophyll-well wavelengths,
and , respectively, from red edge inverted-gaussian
curve fitting [18], as well as spectral indices calcu-
lated from derivative analysis:
; DPR1
, DPR2 , DP21
and DP22 , where
is the value of the reflectance derivative at the spectral
wavelength.
The successful application of such extensive research on
leaf-level optical indices to earth observing instruments at much
broader scales in order to predict canopy condition, requires
the development of links between the leaf and the canopy,
where photon-vegetation interactions are affected by the two
different media. The estimation of pigment content at a canopy
level can be performed using simple statistical relationships at
a leaf level through the use of optical indices [19]–[23], using
modeling methods through radiative transfer by numerical
model inversion [24]–[29] and by a combination of leaf-level
empirical relationships coupled with a canopy reflectance (CR)
model [16], [17]. Research on the application of radiative
transfer models for coupling leaf and canopy models shows
promising results in the simulation of the pigment effect on leaf
reflectance and in turn the effect of the geometrical arrangement
of leaves on the canopy reflectance. Such developments have
the potential to replace the statistically-based approaches for
estimation of leaf bioindicators with quantitative model-based
methods. The application of such methods in forestry canopies,
where canopy structure plays an important role, the selection
of the merit function used in the optimization of simulated
canopy reflectance coupling a leaf and a canopy model, and
the effect of leaf area index (LAI), shadows an understorey
in the modeled reflectance, and therefore, in the estimated
pigment content, we need continued extensive research with
real airborne or satellite-level data. Throughout this paper, the
term LAI represents effective leaf area index
, as defined
in [30], since
can be conveniently defined in terms of
canopy gap fraction. Inherently, this usage ignores the effects
of woody material and foliar clumping needed for a more
detailed specification of LAI [31], but for flat deciduous leaves,
this simplification is considered acceptable. Methodologies
for the application of radiative transfer theory and modeling
assumptions at laboratory and field scales in order to define
the link between leaf reflectance and transmittance and canopy
airborne hyperspectral data acquired with the compact airborne
spectrographic imager (CASI) are discussed in the following
sections. Airborne data acquired with different spectral and
spatial characteristics over twelve Acer saccharum M (sugar
maple) study sites in four consecutive years at 1 m, 2 m, and
3 m spatial resolutions and in 72 spectral channels in the
visible and NIR facilitated the investigation of such important
questions through leaf and canopy radiative transfer models for
estimation.
II. M
ETHODOLOGIES FOR ESTIMATION OF PIGMENT CONTENT
IN
VEGETATION
CANOPIES:SCALING UP AND MODEL
INVERSION
Predictions of chlorophyll content or any other canopy bio-
physical parameter from airborne or satellite canopy reflectance
can been carried out through four different methodologies:
1) directly studying the statistical relationships between
ground-measured biochemical data and canopy-measured
reflectance [19], [20]; 2) applying the leaf-level relationships
derived between optical indices and the pigment content
directly to canopy-measured reflectance [21]–[23]; 3) scaling
up the leaf-level relationships based on optical indices related
to pigment content through models of canopy reflectance or
infinite reflectance
[8], [15]–[17]; and 4) inverting the
observed canopy reflectance through a canopy reflectance or
infinite reflectance model coupled with a leaf model to estimate
the optimum pigment content [17], [24]–[29].
The four proposed methodologies have advantages and dis-
advantages that are related to the complexity of the modeling
approach selected and the degree of general or local applica-
bility of the methodology in remote sensing. The first method
studies the correlations between canopy-measured reflectance
by a field, airborne or satellite sensor with ground-measured
pigment, or any other biophysical constituent. In this case, no
leaf reflectance is measured, and therefore, the link between
canopy reflectance and biochemical content is found through
statistical relationships. Multivariate analysis between visible
infrared imaging spectrometer (AVIRIS) reflectance and total
nitrogen, lignin, starch, chlorophyll content, and LAI [19]
and with nitrogen and chlorophyll [20] applied by stepwise
multiple-regression procedure using the AVIRIS spectral
bands showed good statistical relationships derived at specific
wavebands. Although significant correlations were found, no
predictive capabilities could be inferred to other study sites
since the locally-derived relationships are affected by species
and canopy structure.
The second method, which uses statistical leaf-level relation-
ships applied to canopy reflectance for pigment estimation, is
also site and species specific [32], [33] and therefore requires
relationship calibration that is a function of the canopy struc-
ture and viewing geometry at the time of remote sensing data
collection. Therefore, the statistical relationships derived at leaf
level need to be “calibrated” in order to be useful for estima-
tion at the canopy level, due to the differences between the two
media: one where the relationship is derived (leaf) and the other
where it is applied for estimations (forest canopy). This method-
ology allows the derivation of relationships based on optical in-
dices calculated at wavelengths where subtle changes in leaf re-
flectance correspond to specific biophysical processes that are
targeted for measurement at the canopy level. Stepwise mul-
tiple-regression is often used to develop predictive algorithms
from leaf reflectance , which are then applied to airborne data:
AVIRIS [23] and airborne imaging spectrometer (AIS) spectra
[21]. Laboratory canopy studies [22] and those using AVIRIS
spectra [34] were directed to the identification of spectral bands
at both leaf and canopylevelswhich are less sensitive to changes
between levels, thereby minimizing effects due to the canopy,
ZARCO-TEJADA et al.: SCALING-UP AND MODEL INVERSION METHODS 1493
thus selecting spectral bands that could be used directly for pre-
diction at canopy level. Application of leaf-level relationships
to canopy reflectance through optical indices has been the tra-
ditional method used in the past, and a summary of the optical
indices derived at the leaf level was described in the introduc-
tion.
In the third methodology, the same relationships between
leaf constituent content and canopy reflectance are derived
by scaling up the optical indices through infinite or canopy
reflectance models [8], [9], [17]. A primary advantage is that
the use of infinite or canopy reflectance models as part of the
calculation of relationships avoids the post-calibration step to
compensate for canopy structure or viewing geometry. There-
fore, scaled-up leaf-level relationships can be used directly
for bioindicator predictions on measured canopy reflectance
data by considering canopy structure and viewing geometry
information in the model scaling-up step. The objective of this
method is the derivation of predictive algorithms to be used
under certain canopy assumptions, not simply the evaluation of
statistical correlations between sensor reflectance and ground
measurements. In closed dense vegetation canopies, the re-
flectance canopy model can sometimes be replaced by different
infinite reflectance formulations, as explained later, therefore
simplifying the need for input parameters defining structure and
geometry. As in the first methodology, this approach enables
a search for subtle changes in leaf reflectance due to specific
biophysical processes, and the reflectance model permits direct
prediction of the canopy biochemical parameter. The main
disadvantage is the requirement for leaf sample collection for
the derivation of leaf-level relationships.
The fourth methodology, inversion of a canopy reflectance
model coupled with a leaf model, attempts to avoid the develop-
ment of leaf-level relationships through the use of a leaf model.
In this approach [24]–[27], [29], the leaf radiative transfer simu-
lation uses leaf biochemical constituents as input to model leaf
reflectance and transmittance that is in turn used as input for
the canopy reflectance model. The main advantage of this ap-
proach is that no leaf sample collection is needed to derive rela-
tionships, but suffers from the constraint that only biophysical
parameters considered in the leaf model can be estimated from
measured canopy reflectance. No subtle changes due to spe-
cific functioning effectscan be sought, and therefore no changes
at specific absorption wavelengths due to chlorophyll degrada-
tion at different senescence stages can be studied. That is, it is
implicitly assumed that the leaf model captures all actual ra-
diative processes accurately. Furthermore, the method is com-
putational intensive, and no validation has been found in the
literature reporting results in forest canopies from airborne or
satellite-measured reflectance with ground truth; previous work
has focused on synthetic data [24], [28], field spectrometer data
[25], estimation of results with no validation due to a lack of
ground truth [26], comparison of differentmodel inversion tech-
niques using simulated data [35], and simulation studies mod-
eling three-dimensional (3-D) canopies used for applying in-
version techniques [27], [28]. Successful
and LAI es-
timation results were reported for agricultural crops observed
with airborne CASI data in which comparisons of inversions
from four radiative transfer models coupled to PROSPECT [29],
using Minolta SPAD-502 for estimation of leaf
, are de-
scribed. Other studies with lower spatial and spectral resolu-
tion data [36], [37] use canopy model inversion for extracting
canopy biophysical information from large swath satellite data
at global scales using the advanced very high resolution ra-
diometer (AVHRR) and VEGETATION/SPOT4, respectively,
and therefore, its applicability and portability to narrow-band
hyperspectral high-spatial airborne data cannot be evaluated.
Simulation of the tree crown reflectance spectral content for
comparison to the measured canopy reflectance and retrieved
optical indices may be done through
and CR models,
depending upon the complexity and assumptions made with
respect to the type of vegetation canopy and viewing geometry.
Infinite reflectance formulations model reflectance without
canopy structure or viewing geometry considerations, based
solely on leaf reflectance and transmittance. These formula-
tions correspond to optically thick leaf material with different
assumptions for the multiple scattering between leaf layers.
This thick-leaf or leaf-stack concept may have applicability
to closed deciduous canopies characterized generally by
high LAI, therefore with little effect of soil background and
understorey. CR models, such as SAILH [38] and MCRM
[39], [40] used in this research, on the other hand, take into
account viewing geometry and canopy structure, therefore
modeling those effects in the canopy reflectance by different
approximations generally based on the RTE and geometrical
optical considerations.
Different infinite reflectance formulations have been derived
based on assumptions related to the scattering between layered
leaves forming the optically thick canopy. In each case, the
reflectance for an optically thick medium is expressed in
terms of the inherent single leaf reflectance and transmittance.
Lillestaeter [41] derived a simple formulation (referred to here
as
) from measurements of leaf-stack apparent reflectance
over known dark and bright backgrounds, ignoring multiple
scattering, and considering equal reflectance for both sides of
the leaf (1a). This simple formulation was found inadequate by
Miller et al. [42] to simulate the measured reflectance of leaf
stacks. A matrix formulation by Yamada and Fujimura [43]
was used in a simulation that included multiple reflectance
between leaves and considering different adaxial and abaxial
reflectance for the leaf (
, (1b). The Hapke [44] infinite
reflectance formulae
corresponds to a medium with a
single scattering albedo
assumed approximately equal to
reflectance
transmittance for a pile of leaves, (1c).
The corresponding formulae approximating thick leaf canopies
are
approx. leaf stack (1a)
leaf stack (1b)
thick leaf (1c)
Both CR and
models have been used in this research for
estimation from airborne hyperspectral data collected
over closed Acer saccharum M. forest canopies using the
1494 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 39, NO. 7, JULY 2001
Fig. 1. Schematic view of the overall analysis methodology followed for the
scaling-up method. Leaf-level reflectance and transmittance measurements are
scaled-up to canopy level through infinite and CR models and input parameters
related to the canopy structure and viewing geometry. Relationships between
optical indices calculated from above-canopy simulated reflectance and
ground-truth bioindicators are applied to above-canopy hyperspectral CASI
reflectance to obtain bioindicator estimation. Assessment is made comparing
ground-truth measured with estimated bioindicators.
scaling-up and numerical model inversion methods explained
below.
A. Parameter Estimation by Scaling-Up Optical Indices
A schematic view of the approach used here for estimation
of leaf bioindicators through the scaling-up of leaf-level optical
indices is shown in Fig. 1. Single leaf reflectance and trans-
mittance measurements
from field data sampling are
used for the simulation of above-canopy reflectance through
infinite reflectance and canopy reflectance models, constrained
by a specific canopy model parameter assumption set
.
A specific set of assumed input parameters to the CR model
defines the canopy structure, by a more or less complex set
of canopy parameters, and the viewing geometry, defined by
the solar zenith and azimuth, and viewing angles, needed for
simulating canopy reflectance from single leaf reflectance
and transmittance measurements. Canopy spectral reflectances
(denoted
), or more precisely the above-canopy spectral
bidirectional reflectance factor, simulated through the canopy
simulation model are used to calculate specific optical indices
(denoted ). For a given optical index, a set of values
are calculated from the leaf-level spectral measurements used
for CR simulation. Leaf bioindicators (denoted
) measured
in each leaf sample (
, carotenoids, etc) are used to derive
relationships with the optical indices
calculated from the
above-canopy simulated spectra. Therefore, the relationship
between a given bioindicator (e.g.,
g cm ) and a
given optical index (e.g.,
) is calculated from simu-
lated canopy reflectance rather than from leaf-level measured
reflectance. This relationship is therefore affected by canopy
structural parameters and viewing geometry, which permits its
application to above-canopy measured reflectance. Thus, the
relationships between the set of optical indices
and the
set of bioindicators
are then applied to hyperspectral CASI
reflectance data to obtain bioindicator estimations. To do so,
above-canopy measured CASI reflectance is used to calculate
the CASI-optical indices input using relationships of the form
( ), e.g., g cm ,
with
and constant parameters for above-canopy simulation,
and
an optical index derived from the above-canopy
reflectance. This methodology enables the direct application
of sensor-derived optical indices in scaled-up algorithms that
are therefore a function of the canopy structure and viewing
geometry, precluding the need for calibration of prediction
relationships. Assessment of optical indices as estimators
of bioindicators is then made comparing in-field measured
bioindicators (measured
) with CASI-derived estimations
(estimated
).
B. Parameter Estimation by Model Inversion
The estimation of a biophysical canopy parameter by numer-
ical model inversion can generally be carried out by different
methodologies: 1) look-up tables (LUT); 2) iterative optimiza-
tion (OPT); and 3) neural networks (NNT). The look-up table
technique is conceptually the simplest [35] and consists of the
generation of an output table for a discrete set of input param-
eters covering the expected range of the parameters. The table
is used to find the measured value that is directly related to a
given set of input parameters. This method requires the gen-
eration of large number of cases that are subsequently used to
compare with measured data. Iterative Optimization is the clas-
sical technique for inverting radiative transfer models in remote
sensing [25], [26], [45], [46] and consists of minimizing a func-
tion that calculates the root mean square error (RMSE) between
the measured and estimated quantities by successive input pa-
rameter iteration. Neural networksare nonphysical methods that
relate a set of input variables to a set of output variables by a
learning process and have been shown to be efficient in inver-
sion of canopy models [47], [48].
Iterative-optimization numerical model-inversion techniques
to estimate chlorophyll content using a coupled leaf model and a
canopy model requires three consecutive steps: 1) estimation of
leaf reflectance and transmittance
from a set of leaf model
input parameters such as the parameter to be estimated,
,
and other leaf cellular structural or scattering parameters; 2) es-
timation of canopy reflectance from leaf-level, model-estimated
, , and set of canopy model parameters that define canopy
structure and viewing geometry; and 3) error calculation by
comparison of estimated canopy reflectance
to the at-sensor
measured reflectance
. Error calculation consists in deter-
mining the set of parameters
,
which minimizes a merit function
over the whole spectrum
(2)
where
is the measured canopy spectral reflectance, and
is the modeled canopy spectral reflectance with a set
of
parameters. Different merit functions have been defined
in the literature, each based on different assumptions. The mini-
mizing functionfor numerical model inversion using reflectance
datain severalspectral bandscan be calculated1) from single re-
flectance channels, comparing the estimated with the measured
reflectance in all spectral bands [(2), [24], [25]; 2) using
ZARCO-TEJADA et al.: SCALING-UP AND MODEL INVERSION METHODS 1495
weighting factors that representthe weight givento the th wave-
length. The usual protocol is to choose weighting coefficients
to be proportional to the inverse of the measured canopy re-
flectance
, thereby placing more weight to wave-
lengths in the visible part of the spectrum where pigment ab-
sorption is maximum, and minimizing the impact of errors be-
tween measured and estimated reflectance in the NIR, where
chlorophyll absorption decreases and reflectance is driven by
canopy structure; 3) by a more sophisticated construction of
merit functions [26], where penalization to the merit function
[49] is introduced if the best fit is found when a parameter
being inverted falls outside the prior-established range of al-
lowed values; and 4) building merit functions based on spectral
transforms or vegetation indices [46], in which the merit func-
tion generated is based on the optical index that is supposed to
be related with the parameter subject to estimation, in this case
.
As an example, (3) presents a merit function when the
red-edge spectral parameter
is used for pigment
estimation, which could easily be modified if a combination of
optical indices is used
(3)
where
is the optical index calculated from mea-
sured canopy reflectance, and
is the optical
index calculated from modeled canopy reflectance for a given
set of input parameters
. The use of optical indices in the merit
function has not been found reported in any of the validation
work found in the literature, in spite of the significant inherent
potential of this approach for remote sensing applications. Re-
flectance values measured from airborne or satellite sensors are
a function of illumination, canopy structure, and atmospheric
condition at the time of data collection. On the other hand, esti-
mation of biophysical parameters through optical indices maxi-
mizes the sensitivity to such biophysical parameters, while nor-
malizing external effects due to atmosphere, illumination con-
ditions, and viewing geometry [50], [51]. Therefore in this re-
search, leaf-level optical indices and ratios that showed good
correlation with pigment content are proposed here to be used as
a basis for the merit function for model inversion, as discussed
later.
These different approaches have been tested in this research,
in order to compare the pigment estimation by different tech-
niques using hyperspectral airborne data collected in 1998,
1999, and 2000. This data set provides a valuable validation
database for model inversion with hyperspectral data in closed
dense maple canopies. The experimental methods and materials
used to carry out the pigment investigations are described
below.
III. M
ETHODS AND EXPERIMENTAL DATA
CASI airborne hyperspectral data were collected in deploy-
ments over 12 sites of Acer saccharum M. in the Algoma Re-
gion, ON, Canada, in 1997, 1998, and 1999. A validation of the
methodologies developed with 1997, 1998 and 1999 data over
the 12 study sites was carried out in June 2000 selecting a dif-
ferent set of 14 plots of the same species. The above-canopy
data acquisition using the CASI sensor was divided into three
missions, each with a specific sensor mode of operation: the
Mapping Mission, with 0.5 m spatial resolution and seven spec-
tral bands (Fig. 2); the Hyperspectral Mission, with 2
4m
spatial resolution, 72 channels and 7.5 nm spectral resolution
(Fig. 3), and the Full-Spectral Hyperspectral Mission, with 288
channels and 2.5 nm spectral resolution. The 12-bit radiometric
resolution data collected by CASI were processed to at-sensor
radiance using calibration coefficients derived in the labora-
tory by the Centre for Research in Earth and Space Technology
(CRESTech). Aerosol optical depth data at 340, 380, 440, 500,
670, 870, and 1020 nm were collected using a Micro-Tops III
sunphotometer in the study area at the time of data acquisition
in order to derive aerosol optical depth at 550 nm to be used
to process image data to ground-reflectance using the CAM5S
atmospheric correction model [52]. Reflectance data were geo-
referenced using GPS data collected onboard the aircraft. Final
registration of the hyperspectral mode imagery was achieved by
registration to the CASI mapping mission imagery using visual
identification of ground-referenced 1 m white targets, which
served to accurately identify the location of the sites.
Mean reflectance values per plot were calculated from the
hyperspectral imagery in each Acer saccharum M. study site
of 20
20 m. The mean reflectance per plot was calculated
selecting the 25% of pixels with highest reflectances in the
NIR, therefore targeting crowns while minimizing the influence
of shadows, canopy openings and the direct understorey re-
flectance. The study sites of sugar maple were selected in 1997
from existing provincial plot networks in the Algoma Region,
representing a range of productivity and decline. In particular,
six permanent sample plots from the provincial Growth and
Yield Program [53], [54] were chosen to investigate the effects
of stand productivity in maple. Another six plots were selected
from the provincial Hardwood Forest Health Network [55],
[56] to represent a gradient in maple forest decline. Detailed
stand records exist and these sites are considered representative
of tolerant hardwood forests in the Algoma Region.
A field sampling campaign was carried out for biochemical
analysis of leaf chlorophyll concentration, along with leaf
reflectance and transmittance within the same period of the
field data acquisition. Samplings were carried out in June and
July of 1998 and 1999, and in June 2000, collecting from the
top of the crowns at each one of the twelve Sugar Maple study
sites. Four leaves per tree with five trees per study site were
sampled for measurements of
and spectral measure-
ments of reflectance and transmittance, collecting a total of
440 leaf samples per year. Pigment content measurements from
the leaves were made as in [15], [16]. Biochemical analysis of
samples from 2000 showed a narrower range of
content
compared to the 1998 and 1999 sites, with values falling into
the 29.8–42.7
g cm interval (while in previous years, ranges
were 19.1–41.1
g cm in 1998, and 26.6–45.8 g cm in
1999). LAI measurements were acquired for all the plots using
a PCA Li-Cor 2000 instrument.
Single leaf reflectance and transmittance measurements were
acquired on all leaf samples using a Li-Cor 1800–12 integrating
