OR
nutrient
imics of
, yet our
imaging
ts over a
1 remain
wavelet
e spectra
sis holds
efine the
es. Thus,
nological
mentary
ung and
of plant
itions to
ental /
lities for
'chniques
traction
standard
differing
bsorption
alues at a
letermine
extracts.
d labour-
juantified
s, which
erties of
aves and
reflected
pigments.
S, CASI)
ters (e.g.
solutions
unities to
quantify
y.
nt for the
regetation
vith large
walls and
| allows a
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
large proportion to pass back through the upper epidermis to be
observed as reflected radiation. Pigments, water and other
biochemicals absorb certain wavelengths of radiation which
reduces reflectance in these regions. However, because of the
overlapping absorption features of the pigments, it is difficult to
relate reflectance at a single wavelength to the concentration of
an individual pigment. Furthermore, leaf reflectance can vary
independently of pigment concentrations due to differences in
internal structure, surface characteristics (hairs/waxes) and
moisture content. The reflectance spectrum of a whole canopy
is subject to even more controlling factors, notably, effects of
variations in number of leaf layers (leaf area index; LAI),
orientation of leaves (leaf angle distribution; LAD), presence of
non-leaf elements, areas of shadow and soil/litter surface
reflectance. This range of factors, at leaf and canopy scales,
obscures relationships between spectral reflectance and
concentrations of individual pigments and there has been an
increasing intensity of research aimed at overcoming these
problems. Four groups of spectral variables have been identified
as being of value:
(i) Reflectance in individual narrow wavebands have been
employed (e.g. Fillela er al, 1995). While there is little
agreement on the optimal wavelengths, there is good evidence
that at wavelengths where absorption coefficients of pigments
are high, reflectance is more sensitive to low concentrations,
while spectral regions with low absorption are more sensitive to
higher pigment concentrations (Carter and Knapp, 2001);
(ii) Ratios of reflectance in narrow bands have been proposed as
a means of solving the problems of the overlapping absorption
spectra of different pigments and the effects of leaf structure,
leaf surface interactions and canopy structure (Pefiuclas ef al.,
1995). Most workers propose pigment indices which employ
ratios of narrow bands in the visible and near-infrared (e.g.
Blackburn, 1998a) while some identify only visible
wavelengths and others use combinations of narrow wavebands
in the red edge region (e.g. Tarpley ef al., 2000);
(iii) Characteristics of first and second derivatives of reflectance
spectra have been investigated. It has been suggested that
spectral derivatives have important advantages over spectral
reflectance, such as their ability to reduce variability due to
changes in illumination or soil/litter reflectance. The majority
of workers have used derivatives to define the wavelength
position of the red edge (Are) and illustrated relationships
between Ag and total chlorophyll (Chl for) concentration for
both leaves and canopies. The amplitude of first and second
derivatives of reflectance at particular wavelengths (and
combinations of wavelengths) has also been found to be closely
related to pigment concentrations as has the amplitude of the
first derivative of pseudo absorbance (Blackburn, 1999);
(iv) Measurements of absorption feature depths have been
obtained by fitting a continuum to vegetation reflectance
spectra (Kokaly and Clark, 1999). This approach was extended
by normalising to the band depth at the centre and the area of
the absorption feature and using stepwise regression to identify
optimal combinations of band depths which were used to
estimate accurately Chl /of, a and 5 in dried and ground pine
needles (Curran er al., 2001). :
Most research has focussed on Chls and only recently has
attention been paid to quantifying Cars and anthocyanins from
reflectance spectra, using simple adaptations of the above
approaches (Gitelson ef a/, 2002). Even for Chls, no single
879
spectral approach is emerging as a generic solution. Often
developers of spectral approaches do not test their methods on a
range of vegetation types and this has lead to many species- or
site-specific techniques. Recent literature suggests that of the
spectral approaches that exist, none are sufficiently robust and
remain sensitive to confounding factors such as variations in
chlorophyll fluorescence, leaf surface reflectance, water stress
and specific leaf mass. Moreover, studies testing many spectral
approaches under a range of circumstances have reported a lack
of generality and extendibility (Richardson e/ al, 2002) and
even that hyperspectral approaches offer no improvements over
traditional broadband indices for canopy Ch/ estimation (Broge
and Mortenson, 2002). Indeed, recent work by the author
(Blackburn, 2002) demonstrated limited applicability of
approaches across leaf/canopy/stand scales. Within the same
scale, there was a need for locally derived regression
relationships (e.g. between Agr and Chl rof) and even these were
not transferable between different vegetation types.
Furthermore, papers claiming evidence of robust spectral
approaches (Sims and Gamon, 2002) fail to identify methods to
estimate independently CA/ a and b, or Cars and only
demonstrate convincing results for Chl ror at the leaf scale.
Most research in this field has used individual leaves,
collections of leaves or small plants growing in the laboratory
under controlled conditions. Canopy scale studies have either
derived statistical relationships between ground-measured
pigment data and canopy-measured reflectance, or applied leaf-
scale relationships between optical indices and pigment content
directly to canopy-measured reflectance. Relatively few studies
have examined the applicability of different spectral approaches
as we move from individual leaves to whole plant canopies and
stands. Empirical work by the author on vegetation with a
relatively simple or spatially homogenous canopy architecture
has indicated that some spectral variables are robust predictors
of pigment concentrations from leaf to stand level (Blackburn,
1998b), however, such variables are unsuitable for vegetation
with a more complex structure (Blackburn and Steele, 1999).
Recent work using coupled leaf and canopy radiative transfer
(RT) models has examined the predictive capabilities and
robustness of different spectral approaches for 'quantifying
canopy Chl ror (Haboudane ef al., 2002). While these scaling-
up studies are able to identify spectral indices that are
insensitive to factors such as canopy structure, illumination
geometry and soil/litter reflectance, there is little consensus on
the optimal spectral approaches for estimation of Chl rot. The
numerical inversion of RT models based on measured
reflectance spectra has been used to quantify leaf and canopy
Chl tot (Weiss et al., 2000). Such models afford greater insight
into the underlying functionality of reflectance-based pigment
quantification and the inversion approach promises greater
generality, -however, parameterisation of RT models requires
considerable a priori knowledge of the leaves and canopies
under investigation which can render this approach impractical
for operational use. Nevertheless, a technique that offers greater
potential for extendibility combines the rigour of
(bio)physically-based RT models with the normalising
capabilities and pigment-specificity of a hyperspectral index
which is used as the merit function in the inversion (Zarco-
Tejada ef al., 2001). However, there is a need to substantially
improve the predictive accuracy of this approach and to test it
over a range of vegetation types. In summary, hyperspectral
remote sensing has the potential to satisfy the increasing
demand for information on plant pigments over a range of
spatial scales, yet, a standard analytical approach remains
absent.