You are on page 1of 23

HYPERSPECTRAL REMOTE SENSING

R.N. Sahoo
I.A.R.I., New Delhi -110012
1. Introduction
The term hyperspectral is used to refer to spectra consisting of large number of narrow,
contiguously spaced spectral bands. In the field of remote sensing, the term hyperspectral is
also used interchangeably with other terms such as spectroscopy, spectrometry,
spectroradiometry and rarely ultraspectral. Spectroscopy is a branch of physics concerned
with the production, transmission, measurement and interpretation of electromagnetic spectra.
Spectrometry or spectroradiometry is the measure of photons as a function of wavelength.
Ultraspectral is beyond hyperspectral, a goal that has not been achieved yet. Spectrometers
are used in laboratories, field, aircraft or satellites to measure the reflectance spectra of
natural surfaces. When an image is constructed from an imaging spectrometer that records the
spectra for contiguous image pixels, the terms shift to become imaging spectroscopy, imaging
spectrometry or hyperspectral imaging. Hyperspectral imaging is a new technique for
obtaining a spectrum in each position of a large array of spatial positions so that any one
spectral wavelength can be used to make a recognizable image (Clark, 1999). By analyzing
the spectral features in each pixel, and thus specific chemical bonds in materials, we can
spatially map materials. The narrow spectral channels that constitute hyperspectral sensors
enable the detection of small spectral features that might otherwise be masked within the
broader bands of multi-spectral scanner systems. In this regard, we hypothesis that
hyperspectral sensors could help to overcome the traditional problems faced when using the
broader bands of multi- spectral scanner systems, such as the saturation problem in estimating
quantity and the estimation of quality.
1.1 Advantages of Hyperspectral Remote Sensing
Direct field techniques for estimating soil and vegetation attributes require frequent
destructive sampling. Such techniques are difficult, extremely labour intensive and costly in
terms of time and money. They can hardly be extended to cover large areas. However, remote
sensing technique particularly of high spectral resolution has been found to be very potential
for quantitative assessment of soil and vegetation at spatial scale. A major limitation of
broadband remote sensing products is that they use average spectral information over
broadband widths resulting in loss of critical information available in specific narrow bands
(Blackburn, 1998, Thenkabail et al.. 2000). Recent developments in hyperspectral remote
sensing or imaging spectrometry have provided additional bands within the visible, NIR and
shortwave infrared (SWIR) (Figure 1). Most hyperspectral sensors acquire radiance
information in less than 10 nm bandwidths from the visible to the SWIR (400-2500 nm)
(Asner, 1998). For example, the spectral shift of the red-edge (670-780 nm) slope associated
with leaf chlorophyll content, phenological state and vegetation stress, is not accessible with
broadband sensors (Collins, 1978; Horler, et al.1983).

Figure 1. Comparison of Hyperspectral and multispectral data of a vegetation.


2. Hyperspectral Remote Sensing of Soil
Over the last couple of decades, it has been a challenge to find the most appropriate technique
for studying soil properties effectively and, at the same time, reducing the time and effort
involved in field sampling and laboratory analysis. This has been a major concern not only
for soil scientists but also for environmental specialists. The scope of remote sensing data
has been widely studied, in this respect, looking more closely at how the spectral response of
soil can be linked to various soil properties and characteristics. The relatively high spectral
resolutions as well as contiguous placement of bands covering wider region of electromagnetic spectrum provide more opportunities for soil characterization.
2.1 Hyperspectral Reflectance Spectra for soil characterization
Reflectance spectra have been used for many years as one of the source of information about
the variation in the Earth's surface composition (Van der Meer et al. 2001). Consequently, the
soil properties derived from these spectra have been studied as early as 1980s. Study of soil
properties controlling soil reflectance and conversely observations of soil reflectance
provides information on some of the soil properties and, in general, the state of soil quality
(Irons et al. 1989). Many studies have drawn their main emphasis on the visible and infrared
regions of the electro-magnetic spectrum- between 300 and 2500 nm (Baumgardner et
al.1985).
In general, a wide range of information can be obtained from reflectance properties related to
the nature and chemical composition of the soil material (Stoner and Baumgardner, 1981;
Baumgardner et al. 1985). This is mainly based on specific absorption of spectrally active
groups (known as chromophores), such as Fe, OH- in water and minerals, CO32-, Al2+, (Mg+)
OH, SO42- in minerals, and any others in organic matter (Ben-Dor et al. 1997). Reflectance
data have been successfully used for prediction of numerous soil properties, such as soil
moisture and organic matter content in laboratory studies (Dalal and Henry, 1986; Ben-Dor
and Banin, 1995a). Further, many researchers have studied the effect different factors
affecting soil reflectance and thereby could predict them from soil reflectance.
Spectral reflectance has been widely used by many in the visible and near infrared (VIS/NIR,
400- 2500 nm) as well as in mid infrared (MIR) spectral regions for assessment of soil
chemical fertility. Most common fertility parameters studied were Soil organic carbon
(SOC), inorganic carbon (SIC), total nitrogen (TN), cation exchange capacity (CEC), pH,
potassium (K), magnesium (Mg), calcium (Ca), zinc (Zn), iron (Fe), and manganese (Mn)
with various levels of prediction accuracy (e.g. Baumgardner et al. 1985; Ben-Dor and
Bannin, 1995a; Malley et al. 1999; Shepherd and Walsh, 2002; Viscarra Rossel et al. 2006).

II.108

For characterizing soils from hyperspectral reflectance data Kumar et al. (2006) collected
eighty seven surface soil samples (0-15cm) representing 23 soil series were collected from 23
sites of five different places of India encompassing 5 different climatic zones including soils
from 4 soil taxonomic orders (Vertisols from Nagpur, Alfisols from Ranchi and Inceptisols
from IARI farm) on the basis of reports on soils of different states published by the National
Bureau of Soil Survey and Land Use Planning and spectral studies were carried out using FS 3
ASD spectroradiometer (having spectral range 350-2500nm). Reflectance spectra of soil
samples that represent four soil orders within the examined population are given in Fig.1.
These signatures can easily be attributed to several common soil constituents. The most likely
component in the NIR region is the OH groups in both the adsorbed water (at 1.4m and
1.9m) and the crystal water of clay minerals (at 1.45m and 2.2m ) (Hunt and Salisbury
1970). Also the CO3 in the calcite mineral is very active in the NIR region of soils (major
peak at 2.33) (Hunt et al.1971; Gaffey 1986).
From the analysis of the reflectance spectra pattern different soils orders namely Alfisols,
Inceptisols, Mollisols and Vertisols could also be discriminated. The spectral pattern of
Alfisols showed higher reflectance pattern in the IR region (Fig 1) than the others three soils
orders (Inceptisols, Vertisols, and Mollisols), due to high free iron oxide content as well as
due to high kaolinite and interstratified kaolinite content. Alfisols showed high absorption
pattern in the visible region than the Inceptisols and Vertisols. These soils showed strong
absorption peaks near 450 nm and weak absorption peak near 550 nm. The absorption band
near 450 nm is caused by paired and single Fe3+ electron transitions to higher energy state
(Sherman and Waite,1985), the small absorption peak near 550 nm may be due to the
chromophore FeO-OH found in goethite (Mortimore et al. 2004) which is dominated in
Alfisols. In VIS region, convex curve was found which is characteristic of Alfisols soil
orders. These convex curves were broader for high iron and free iron oxide containing soil
samples. Mollisols spectral reflectance was lowest one as compared to all other soil orders
throughout the spectral region (350-2500 nm). Lowest spectral reflectance pattern of
Mollisols was mainly due to the high organic matter content. Spectral reflectance pattern of
Vertisols showed higher reflectance pattern than Mollisols but lower than Inceptisols and
Alfisols. In VIS region concave curve was observed for Vertisols as well as for Mollisols.
The concave and convex pattern of the reflectance spectra were observed in the VIS region
because these were guided with organic matter and iron content of the soils which were
highly correlated in this region. Spectral reflectance of Inceptisols was observed to be higher
than Vertisols and Mollisols but less than Alfisols. It may be due to its low organic matter
content and sandy loam texture.
VIS (350-710 nm) region of spectra was dominated with several features and specific pattern
with respective bands where but NIR (710-1110 nm) region was featureless. (Fig 1). These
four soil orders namely Mollisols, Vertisols, inceptisols, and Alfisols in red bad (650-700 nm)
showed around 10, 15, 25, and more than 25 % reflectance respectively. So VIS region was
found to be more useful for discriminating these four soil orders as compared to NIR region
of spectra. The spectral pattern in short wave infrared (SWIR, 1100-2500 nm) region of
Alfisols shows the characteristic absorption peak near 1400, 2200, and 2300 nm bands. This
1400 nm peak may be attributed due to the first overtone of the O-H stretch and a second
doublet at 2200 and 2300nm is due to the combination Al-OH bend plus O-H stretch. This
metal-OH bend plus O-H stretch combination near 2200 nm and 2300 nm are diagnostic
absorption features for clay minerals (Clark et al. 1990). Inceptisols spectra in SWIR region
also showed characteristic absorption peaks at near 1400, 2200, and 2300 nm bands. These
two soil orders alongwith its characteristics absorption peaks, also have universal water
absorption band near 1900 nm. Alfisols showed stronger reflectance peaks near 2300 nm, due

II.109

to the presence of surface O-H group as compared to Inceptisols. Inceptisols showed


absorption band near 2200 and 23000 due to additional Al-OH features. Vertisols spectra in
SWIR region showed less pronounced peak near 1400 nm band which is also characteristic
absorption peak for Inceptisols and Alfisols. However Vertisols have characteristic broader
and deeper absorption peaks near 1900 nm band. These characteristics absorption peaks was
observed due to presence of smectite clay mineral where water molecules are absorbed
between the sheets of the smectite structure. Clark et al. 1990 have found that the intense
absorption peaks present in smectite dominated soils is due to the combination of the H-O-H
bend with the O-H stretches. Less pronounced absorption peak was also observed near
2200nm. The spectrum that has a 1400 nm band but not 1900 nm band indicates that only OH group is present and only a small amount of water, because of a weak 1900 nm absorption.
But presence of a intermediate O-H absorption feature at 1400nm is indicative of Inceptisols.
Vertisols were found to have the least O-H absorbtion peak at 1400 nm. It was observed that,
when the amount of kaolinite content decreases, the characteristic absorption bands at 1400
nm and 2200 nm become less pronounced in Alfisols. Conversely as the amount of smectite
increases in soil samples, the characteristic 1900 nm water absorption band becomes more
broader in vertisols. Spectral pattern of Mollisols in SWIR region were observed to be quite
different than rest three soil orders. Mollisols spectra showed absorption band near 1400 nm
and 1900 nm, which was neither intense like Inceptsols and Alfisols nor broader like
Vertiisols. These may be due to lack of surface O-H group and presence of non expanding
type of minerals (Vermiculite, Mica, Chlorite etc).
(a)

(b)

(c)

(d)

Figure 1(A). Spectral reflectance pattern of representative soil of four soil orders in (a)
full range, 350-2500nm, (b) only visible range (c) only NIR range and (d)
SWIR range.

II.110

2.2 Quantitative Estimation of Soil Parameters from Hyperspectral Remote Sensing


Hyperspectral data being of larger volume, overlapping of weak overtones and fundamental
vibrational bands, have been very difficult for its direct interpretation (Wetzel, 1983).
Therefore, multivariate analysis is required for quantitative interpretation of soil parameters
from hyperspectral reflectance data. A number of different calibration techniques are
available and have been applied when relating measured spectra to measured values of soil
properties. The choice of calibration technique will depend on the application of the data.
Principal components regression (PCR) and stepwise multiple-linear regression are the most
common (Wise et al. 2003). Different pre-processing transformations have been applied in
numerous studies to transform soil spectral data, remove noise, accentuate features, and
prepare them for chemometric modelling. Pre-processing transformations of spectral data
constitute an important step in multivariate calibration and have been shown to improve the
accuracy of prediction models (Dunn et al. 2002; McCarty et al. 2002).
2.2.1 Multivariate regression approaches
Linear Regression estimates the coefficients of the linear equation, involving one or more
independent variables that could best predict the value of the dependent variable (Gomez and
Gomez, 1984). The complexity and multicollinearity are two common properties inherent to
hyperspectral data. On the other hand, in view of weak correlations of individual reflectance
data with most of the soil properties, univariate modelling is rarely applicable for
hyperspectral use of soil properties. Dardenne (1996) used two multivariate regression
techniques namely multiple linear regression (MLR) and partial least-square regression
(PLSR) to estimate soil parameters from the hyperspectral image data.
Kumar et al. (2009) evaluated the ability of ASD hyperspectral data for estimating 17 soil
properties such as mineralizable nitrogen(N), available phosphorous (P) and potassium (K),
DTPA extractable manganese (Mn), iron (Fe), copper (Cu) and zinc (Zn), calcium carbonate
(CaCO3), soil organic carbon (SOC), pH (1:2.5), EC (1:2.5), bulk density (BD), particle
density (PD), hydraulic conductivity (Ks) and soil texture. The spectral data of 85 soils of
Jalandhar district of Punjab was recorded both at field and laboratory conditions (Fig. 2).
Spectra sensitivity analysis was done to find out best spectral range for different soil
parameter. Example soil organic carbon is given in Fig. 3. Step wise regression approach
was used and results revealed that models developed using derivative spectral data were able
to predict some selected parameters with reasonable higher accuracy while reflectance and
absorbance values did yield poor results. Based on adjustable R2 of the predicted models, first
derivative of absorbance was found suitable for nitrogen while its second derivative was best
for Mn, Fe, and Zn for prediction model. Second derivative of reflectance was selected for P
and Cu and First derivative of reflectance was better for K prediction. The highest
predictability (adjusted R2) was 0.93 recorded for CaCO3 while lowest 0.68 was for N. The
comparison of predicted and measured values of some soil parameters are given in Fig. 4.

II.111

Figure 2. Study area and soil spectral data collection in the farmers field in situ and
laboratory condition.

Figure 3. Spectral sensitivity analysis of Soil organic Carbon.

II.112

Figure 4. Evaluation of Predicted models for some soil properties


2.2.2 Principal component analysis
Principal component analysis (PCA) involves a mathematical procedure that transforms a
number of possibly correlated variables into a smaller number of uncorrelated variables
called principal components. The first principal component accounts for as much of the
variability in the data as possible, and each succeeding component accounts for as much of
the remaining variability as possible. PCA involves the calculation of the eigen value
decomposition of a data covariance matrix or singular value decomposition of a data matrix
(SAS Inst., 1999). Now it is mostly used as a tool in exploratory data analysis and for
making predictive models with hyperspectral data (Campbell, 1996; Ehsani, et al. 1999).

II.113

2.2.3 Continuum removal


The continuum removal approach aims at quantifying the absorption by materials at a
specific wavelength assuming that no other material has strong absorption features around
this specific wavelength (Clark and Roush, 1984). The continuum is approximated by a
straight line joining the two local reflectance maxima placed on both shoulders (min and max)
of the peak absorption wavelength (peak). Continuum removal, CR, was thus written as a
function of reflectance values R() at wavelength , with the constraint that its maximum
value could not be above 1.0 (concavity of the reflectance spectra at this location). Specific
absorption peaks for different chemical components mixed with soil can be applied for
variability assessment (Chabrillat et al. 2002).
Santral et al. (2009) could retrieve spectral transformation functions (STF) from measured
soil reflectance data to predict some of most difficult soil parameters e.g. Van Genuchten
Parameters ( and n) which is used to derive soil hydraulic conductivity. The spectral data
was collected using FS3 ASD spectroradiometer and integrated band reflectance
corresponding to ETM+ of Landsat-7, computed CR factor were used to develop the STFs
and predict soil hydraulic parameters. Comparison predicted and observed of value of soil
hydraulic parameters are given in Figure 5.

Figure 5. Observed and predicted values of saturated hydraulic conductivity [ln(Ks)] (a and
b), and van Genuchten parameters, ln() (c and d), and van Genuchten parameter,
n (e and f) with the best performances (left side) and worst performances (right
side) among pedotransfer functions and spectrotransfer functions developed with
different combination of predictor variables for each hydraulic property.

II.114

2.3 Quantification of soil properties from imaging spectroscopy


Hyperspectral data analysis can be divided into two categories (Aspinall et al. 2002). One is
top-down, which essentially uses field maps to train the imagery so as to detect certain
features of the examined objects. Field survey and geo-referencing are necessary because
high positioning accuracy is needed in this approach. Atmospheric correction of the imagery
may not be needed. Since the high spatial resolution image usually does not have large extent
coverage, the atmospheric variation within each scene is not noticeable. Associating the
ground features directly with the image enables the classifying algorithm to incorporate
atmospheric effects into the feature spectra to search for similar features on the same image.
However, this approach was not found feasible for large physical extent analysis (Aspinall et
al. 2002). The other approach, in contrast, is bottom-up which typically uses ground- or
laboratory-measured spectral libraries to identify features in the image. Atmospheric
correction is essential because atmospheric effects need to be removed from the image before
it can be quantitatively compared with ground-measured spectra. Geo-referencing and
registration are necessary to match geographic positions of image pixels to ground sites. To
reduce spectral noise, the absorption regions affected by water and carbon-dioxide of the
spectrum should be removed before further processing (Kruse, 1994).
Kadupitiya et al. (2009) evaluated Hyperion sensor of EO-1 for estimation of eight soil
properties such as soil organic content (SOC), CaCO3, Mineralizable Nitrogen (N), available
Phosphorous (P), Potassium (K), sand %, silt % and clay % in the Jalandhar district of Punjab
and compared with the proximal reflectance data (through ASD FS3) collected synchronizing
with the satellite pass. The study showed that, the Hyperion spectral pattern were comparable
with ground measured spectra after atmospheric correction using FLAASH (Fig. 6). Derived
spectral parameter such as first and second derivative of reflectance and absorbance were
found to better suited than original reflectance data for developing prediction models for soil
properties irrespective of the platform of the sensor. Comparison of hyperion data with
proximal reflectance data of spectroradiometer revealed that R2 of prediction models
decreases when we move from using laboratory spectra to field and then to Hyperion sensor.
However, some parameters like SOC and CaCO3 could be predicted with reasonable
accuracy from Hyperion sensor (Fig. 6).

II.115

Figure 6. Comparison of Hyperion reflectance data with field and laboratory reflectance
spectra. Brown and green line is soil reflectance collected in laboratory and field
conditions respectively and blue line for hyperion data. Light blue strip in the
graphs indicates the spectral range not considered due to noise.

Figure 7. Comparison of adjusted R2 of prediction models developed from laboratory, field


and Hyperion sensor spectra

II.116

3. Hyperspectral Remote Sensing of Vegetation


Leaves represent the main surfaces of plant canopies where energy and gas are exchanged.
Hence, knowledge of their optical properties is essential to understand the transport of
photons within vegetation (Despan and Jacquemound, 2004). The general shape of
reflectance and transmittance curves for green leaves is similar for all species. It is controlled
by absorption features of specific molecules and the cellular structure of the leaf tissue (Ustin
et al.1999). Three spectral domains can be distinguished (Fig 8): In the visible domain (400700 nm) absorption by leaf pigments is the most important process leading to low reflectance
and transmittance values. The main light absorbing pigments are chlorophyll a and b,
carotenoids, xanthophylls, and polyphenols. Chlorophyll a is the major pigment of higher
plants and together with chlorophyll b account for 65 percent of the total pigments.
Chlorophyll a and b have absorption bands in the blue at around 430/450 nm and in the red
domain at around 660/640 nm. These strong absorption bands induce a reflectance peak in
the green domain at about 550 nm. Carotenoids and xanthophylls absorb mainly in the blue
and are responsible for the colour of flowers, fruits, and the yellow colour of leaves in
autumn. Polyphenols (brown pigments) absorb with decreasing intensity from the blue to the
red and appear when the leaf is dead (Verdebout et al. 1994).
In the near-infrared domain (near-IR: 700-1300 nm) leaf pigments and cellulose are almost
transparent, so that absorption is very low and reflectance and transmittance reach their
maximum values. The level of reflectance on the near-IR plateau increases with increasing
number of intercell spaces, cell layers, and cell size. Scattering occurs mainly due to multiple
refractions and reflections at the boundary between hydrated cellular walls and air spaces
(Guyot, 1990).
In the mid-infrared domain (mid-IR: 1300-2500 nm), also called shortwave-infrared (SWIR),
leaf optical properties are mainly affected by water and other foliar constituents. The major
water absorption bands occur at 1450, 1940, and 2700 nm and secondary features at 960,
1120, 1540, 1670, and 2200 nm (Ustin et al. 1999). Water largely influences the overall
reflectance in the mid-IR domain and also has an indirect effect on the visible and near-IR
reflectances. Protein, cellulose, lignin, and starch also influence leaf reflectance in the midIR. However, the absorption peaks of those organic substances are rather weak as they result
from overtones or combinations related to fundamental molecular absorptions in the region 5
to 8 m (Curran, 1989; Table 1). The molecular absorptions are associated with certain
chemical bonds, such as C-H, N-H, C-O, and O-H. In fresh leaves, spectral features related to
organic substances are masked by the leaf water, so that estimation of leaf constituents is
difficult (Verdebout et al.1994).

II.117

Figure 8. Typical Reflectance Curve of Vegetation and causes of spectral characteristics


(Jensen, 2000).

II.118

Table 1: Absorption features in vegetation reflectance spectra related to leaf nitrogen


concentration (Adapted from Curran, 1989; Lucas and Curran, 1999)

The spectral properties of live foliage set up the radiation field in a canopy, and these spectral
properties express the presence and abundance of both the inputs and products of
photosynthesis (Figure 9). Leaf pigments such as chlorophylls, carotenoids, and anthocyanins
are expressed in the 400-700 nm range, matching the wavelength region of maximum solar
input to the biosphere. The relative contribution of pigments to the reflectance and
transmittance properties of foliage varies by wavelength in this region, and all pigments have
overlapping absorption features (Figure 10). Chlorophyll a (chl-a) displays maximum
absorptions in the 410- 430 nm and 600-690 nm regions; whereas chlorophyll b (chl-b) shows
maximum absorptions in the 450-470 nm range. Carotenoids absorb most efficiently between
440 and 480 nm. In the foliage of many canopy species, chl-b dominates the overall
absorption spectrum at shorter and longer wavelengths in the visible spectrum, whereas
carotenoids can be a major contributor at slightly longer wavelengths (gray line, Figure 10).
In the near-infrared (700-1300 nm) and shortwave-infrared (1300-2500 nm), the leaf
spectrum is dominated by water content, thickness, and, to a lesser degree, protein-nitrogen
(N), cellulose and lignin content (Figure 9) (Curran 1989). In particular, live foliage is a very
efficient at scattering radiation in the 750-1300 nm range; this is caused by internal scattering
at the air-cell-water interfaces within the leaves (Thomas et al.. 1971, Hunt et al.. 1987). At
longer wavelengths (> 1300 nm), relatively small amounts of water (and resulting air-water
interfaces) 4 effectively trap radiation, resulting in absorption that exceeds scattering
processes in this portion of the spectrum. Meanwhile, leaf structural properties, especially
area per mass (specific leaf area or SLA), play an underlying but integral role in containing
the biochemicals in a functional form adapted for leaf carbon fixation and related
photosynthetic processes (Jacquemound et al..1996, 2000).

II.119

Figure 9. The radiance spectrum of a leaf or canopy expresses the presence and
abundance of the building blocks and products of photosynthesis. Elemental
and molecular contributions to the spectrum are labeled. This spectrum was
acquired over lowland Hawaiian rainforest using the Airborne Visible and
Infrared Imaging Spectrometer (AVIRIS).

II.120

Figure 10. Relative absorption intensity of chlorophyll-a and b, as well as bulk


carotenoids, in Metrosideros polymorpha
What is the contribution of leaf N to reflectance signatures of tropical forests? Proteins have
absorptions spread throughout the shortwave-IR spectrum (Curran 1989) that are partially
obscured by water absorptions. In addition, chlorophyll is ~6% nitrogen by mass, and thus
chlorophyll and N tend to be broadly correlated with one another. Remote estimation of
chlorophyll tends to scale with N, with a range of correlation coefficients (r) from about 0.7
to 0.9 (Yoder and Pettigrew-Crosby 1995). Moreover, spectral analyses between leaf or
canopy spectra and nitrogen often show spectral correlations in the wavelength regions
associated with
both chlorophyll and protein-N (Martin and Aber 1997, Kokaly 2001, Smith et al.. 2003). But
are
correlations between spectra and N concentration direct, or are they indirect by way of leaf
chlorophyll and other leaf constituents? Leaf construction involves a stoichiometric balance
between chlorophyll, nitrogen, water and other biochemicals, all convolved with leaf
structure, especially SLA (Reich et al.. 1997, Evans and Poorter 2001). Leaf protein is
dominated by the enzyme Rubisco, which mediates carbon fixation in concert with light
capture by chlorophyll (Niinemets and Tenhunen 1997, Stylinski et al. 2000). Leaf water
concentrations scale with SLA, and SLA is correlated with leaf N per unit area (Wright et al.
2004). These leaf parameters are therefore not independent, biophysically or ecologically,
and thus the retrieval of leaf N tends to be wrapped up in the retrieval of the other parameters.

II.121

3.1 Spectral Transformations and parameter estimation

3.1.1 First derivative reflectance


A first difference transformation of the reflectance spectrum (FDR) calculates the slope
values from the reflectance and can be derived from the following equation (Dawson &
Curran, 1998). Derivative spectra have been widely used in spectroscopy and remote sensing.
Reflectance spectra may suffer from background signals and albedo effects. In contrast,
derivatives were shown to be less sensitive to variations of soil background, illumination, and
surface albedo (Demetriades-Shah et al.1990). The main disadvantage of derivatives is their
sensitivity to sensor noise. The decrease in the signal-to-noise ratio (SNR) with increasing
derivative order often limits the usage of higher order derivatives. The wavelength having
highest FDR is called red edge position (REP). If the value of REP shifts towards lower
wavelength, it is called blue shifting and indicates nutrient (nitrogen) stress and shifting
towards higher wavelength is called red shift and indicates healthy condition (Figure 11 a and
b).

(a)

(b)
Figure 11(a and b). Spectral Shift in healthy and stressed plant and (b) REP position for
low/nil and high N treated plot of wheat crop.
3.1.2 Continuum-removal and band-depth normalization
Absorptions in a spectrum are composed of the continuum and individual features, where the
continuum is the background absorption onto which other absorption features are placed
over. The continuum is simply an estimate of the other absorptions present in the spectrum,
not including the one of interest (Clark & Roush, 1984). In an experiment, (Mutanga,
Skidmore et al.. 2003) tested the utility of using four variables derived from continuumremoved absorption features for predicting canopy nitrogen, phosphorus, potassium, calcium
and magnesium concentration (Figure 12).

II.122

Figure 12. Estimation of canopy nitrogen, phosphorus, potassium, calcium and magnesium
concentration using four variables derived from continuum-removed absorption
features (From Mutanga et al.2003)
3.2 Parameter Retrieval from Hyper BRDF through radiative transfer modelling
Different methods to estimate canopy biophysical variables from reflectance data have been
developed and can be grouped into two approaches such as (1) statistical approach and (2)
physical process based approach (using Radiative Transfer Models). Using statistical
approach, many researchers have developed empirical relationships between vegetation
indices (VIs) and canopy biophysical variables. The equations defining such empirical and
semi-empirical relationships not only vary in the mathematical form (Linear, power,
Exponential, etc) but also in their empirical coefficients, depending upon the cultivars,
regions and the data normalizations approaches adopted. These methods are very simple but
the accuracy of biophysical variable estimation may be quite low. They are suffering from
severe limitations due to the lack of physics introduced in the retrieval technique and the
small amount of radiometric information they can exploit. Alternately, physical modeling
approach is based on the inversion of canopy reflectance models that describe the radiative
transfer in the canopy as a function of biophysical variables which characterize the canopy
architecture and the optical properties of vegetation elements and the soil. In the mid-80s, the
anisotropic properties were observed to be crucial for diagnosing plant canopy functioning.
Enhanced understanding of the physical processes that govern the interactions between light
and the canopy elements, bi-directional canopy reflectance (CR) models emerged for its

II.123

inversion issues on multidirectional data in the early 90s for retrieval of biophysical
parameters (Goel, 1987).
Inversion of bidirectional canopy reflectance (CR) models emerged as a promising alternative
for retrieval issues (Goel 1989; Myneni et al. 1991; Liang and Strahlar, 1993; Tripathi et al.
2006, 2009). The space borne instruments like POLDER, ADEOS, MISR, TERRA, etc were
designed to study both the spectral and directional characteristics of the earth surfaces. This
trend depicts one of the scientific stakes to come in remote sensing, which is to take
advantage of both the spectral and the directional signatures of vegetation in order to retrieve
the biophysical parameters.
Tripathi et al. (2006) conducted one field experiment with the objectives set as (i) to relate
canopy biophysical parameters and geometry to its bidirectional reflectance, (ii) to evaluate a
canopy reflectance model to best represent the radiative transfer within the canopy for its
inversion and (iii) to retrieve crop biophysical parameters through inversion of the model.
Two varieties of the mustard crop (Brassica juncea L) were grown with two nitrogen
treatments to generate a wide range of Leaf Area Index (LAI) and biomass. The reflectance
data obtained at 5nm interval for a range of 400-1100nm were integrated to IRS LISS II
sensors four band values using Newton Cotes Integration technique. Biophysical parameters
were estimated synchronizing with the bi-directional reflectance measurements. The radiative
transfer model PROSAIL was used for its evaluation and to retrieve biophysical parameters
mainly LAI and Average Leaf Angle (ALA) through its inversion. Look Up Table (LUT) of
BRDF was prepared simulating through PROSAIL model varying only LAI (0.2 interval
from 1.2 to 5.4 ) and ALA (5 interval from 40 to 55) parameters and inversion was done
using a merit function and numerical optimization technique given by Press et al. 1986. The
derived LAI and ALA values from inversion were well matched with observed one with
RMSE 0.521 and 5.57, respectively.
The radiative transfer model PROSAIL was used for retrieval of LAI, Chlorophyll (Cab) and
equivalent water thickness (Cw) of wheat crop of Trans Gangetic Plains through its inversion
(Fig 13). The model was calibrated for major parameters such as LAI, Cab, Cw and biomass
(Cm) and sensitivity analysis was performed. Inversion of PROSAIL model was carried out
for LAI, Cab and Cw using Look Up Table (LUT) approach. The merit function was
computed and used to best fit the measured data with the simulated one. Results revealed that
LAI, Cab and Cw, were very well retrieved with RMSE 0.3892, 4.307 and 0.0063
respectively when compared with measured values. The retrieved products were evaluated
with its corresponding regressed products through different vegetation indices. RMSE
between these regressed estimation and measured parameter values were 0.553, 5.204 and
0.01 for LAI, Cab and Cw respectively.

II.124

4.
5.
6.
7.
8.
9.
10.
11.
12.

EWT

Chl

Figure 13. The LAI, Equivalent Water Thickness and Total Chlorophyll content of wheat
crop in Trans Gangetic Plain of India retrieved from MODIS data through
radiative transfer modelling.
References
1.

American Society of Photogrammetry, (1968). Manual of Color Aerial Photigraphy,


Falls Church, Va.

2.

American Society of Photogrammetry, (1980). Manual of Photogrammetry, 4th ed.,


Falls Church, Va.

3.

American Society of Photogrammetry, (1983). Manual of Remote Sensing, Vol. I, R. N.


Colwell (Ed.), Falls Church, Va.

4.

Barbe, D.F. (1980). Charge Coupled Devices. Topics in Applied Physics, Vol.38.
Springer- Verlag, Berlin.

5.

Baret, F. and Guyot, G. (1991). Potential and limits of vegetation indices for LAI and
APAR assessment. Remote Sensing of Environment, 35, 161-173.

6.

Barman, D., Sehgal, V.K., Sahoo, R.N. and Nagarajan, S. (2010). Relationship of
Bidirectional Reflectance of Wheat with Biophysical Parameters and its Radiative
Transfer Modeling Using PROSAIL, J. Ind. Soc. Remote Sens., 38, 141- 151.

7.

Boegh, E., Soegaard, H., Broge, N., Hasager, C.B., Jensen, N.O. and Schelde, K.
(2002). Airborne multispectral data for quantifying leaf area index, nitrogen
concentration, and photosynthetic efficiency in agriculture. Remote Sensing of
Environment. 81,179193.

8.

Bonham-Carter, G.F. (1988). Numerical procedures and computer program for fitting an
inverted Gaussian model to vegetation reflectance data. Computer & Geosciences,
14,339-356.

9.

Broge, N.H. and Leblanc, E. (2000). Comparing prediction power and stability of
broadband and hyperspectral vegetation indices for estimation of green leaf area index
and canopy chlorophyll density. Remote Sensing of Environment. 76, 156-172.

10. Carter, G.A.(1994). Ratios of leaf reflectances in narrow wavebands as indicators of


plant stress. International Journal of Remote Sensing, 15, 697704.

II.125

11. Casa,R. and Jones, H.G. (2004). Retrieval of crop canopy properties: a comparison
between model inversion from hyperspectral data and image classification.
International Journal of Remote Sensing, 23, 1119-1130.
12. Cheng, Cheng Ji, (Ed.). (1989). Manual for the Optical-Chemical Processing of
Remotely Sensed Imagery. ESCAP / UNDP Regional Remote Sensing Programme
(RAS/86/141), Beijing, China.
13. Cho & Skidmore (2006). A new technique for extracting the red edge position from
hyperspectral data: The linea.r extrapolation method. Remote Sensing of Environment
101, 181193.
14. Clark, R.N. and Roush, T.L. (1984). Reflectance spectroscopy: Quantitative analysis
techniquesfor remote sensing applications. Journal of Geophysical Research, 89 (B7),
6329-6340.
15. Clark, R.N. (1999). Spectroscopy of rocks and minerals and principles of spectroscopy.
In:Rencz, A. N. (Ed.): Remote sensing for the Earth sciences: Manual of remote
sensing, 3rd ed., Vol. 3 New York, Chichester, Weinheim. John Wiley & Sons, Inc.: 358.
16. Curran, P.J.(1989). Remote sensing of foliar chemistry. Remote Sensing of
Environment, 30, 271-278.
17. Curran, P. J. (1985). Principles of Remote Sensing, Longman, London.
18. Dawson, T.P. and Curran, P.J. (1998). A new technique for interpolating the reflectance
red edge position. International Journal of Remote Sensing, 19(11) ,2133-2139.
19. Demetriades Shah, T.H., Steven, M.D. and Clark, J.A. (1990). High resolution
derivative spectra in remote sensing. Remote Sensing of Environment, 33, 55-64.
20. Despan, D. and Jacquemoud, S. (2004). Optical properties of soil and leaf: Necessity
and problems of modeling. In: Schnermark, M. von, Geiger, B., Rser, H. P. (Eds.):
Reflection properties of vegetation and soil. Berlin. Wissenschaft und Technik Verlag:
39-70.
21. Duckworth, J. (1998). Spectroscopic quantitative analysis. In: Workman, J.,
Springsteen, A. W. (Eds.): Applied spectroscopy. A compact reference for practitioners.
San Diego, New York, London. Academic Press: 93-164.
22. Elachi, Charles, (1987). Introduction to the Physics and Techniques of Remote Sensing,
John Wiley & Sons, New York.
23. Franklin, E. (Eds.): Remote sensing of forest environments - Concepts and case studies.
Boston,Dordrecht, London. Kluwer Academic Publishers: 279-300.
24. Franklin, J., Phinn, R., Woodcock, C.E. and Rogan, J. (2003). Rationale and conceptual
framework for classification approaches to assess forest resources and properties. In:
Wulder, M. A.,
25. Goel, N. (1989). Inversion of canopy reflectance models for estimation of biophysical
parameters from reflectance data, In Theory and Applications of Optical Remote
Sensing (G. Asrar, Ed), Wiley Interscience, New York.205-250.
26. Goel, N.S. (1987). Models of vegetation canopy reflectance and their use in estimation
of biophysical parameters from reflectance data, Remote Sensing Reviews, 1, 221.

II.126

27. Goel, N.S. (1989). Inversion of canopy reflectance models for estimation of biophysical
parameters from reflectance data, In Theory and Applications of Optical Remote
Sensing (G. Asrar, Ed), Wiley Interscience, New York, 205-250.
28. Guyot, G. (1990). Optical properties of vegetation canopies. In: Steven, M. D., Clark, J.
A. (Eds.): Applications of remote sensing in agriculture. London. Butterworths: 19-43.
29. Horler, D.N., Dockray, M. and Barber, J. (1983). The red edge of plant leaf reflectance.
International Journal of Remote Sensing. 4, 273-288.
30. Jacquemoud, S. and Baret, F. (1990). 'PROSPECT: A model of leaf optical properties
spectra, Remote Sensing of Environment. 34, 251-256.
31. Jacquemoud, S. (1993). Inversion of the PROSPECT + SAIL canopy reflectance model
from AVIRIS equivalent spectra: Theoretical study, Remote Sensing of Environment,
44(2-3), 281-292.
32. Jensen, J.R. (1996). Image Preprocessing: Radiometric and Geometric Correction. In :
Introductory Digital Image Processing. Prentice Hall publisher (Second Edition). 107135.
33. Kadupitiya, H.K., Sahoo, R.N., Gupta, V.K., Ahmed, N. and Ray, S.S. ( 2009).
Modeling soil chemical composition using hyperspectral reflectance data., In ISRS
Annual Convention and National Symposium on Advances in Geo-spatial Technologies
with special emphasis on Sustainable Rainfed Agriculture by Indian Soc. Of Remote
Sensing, Nagpur, Sept 17-19.
34. Kadupitiya, H.K., Sahoo, R.N., Ray, S.S., Chakraborty, D. and Ahmed, N. (2010).
Quantitative Assessment of Soil Chemical Properties using Visible (VIS) and Nearinfrared (NIR) Proximal. Hyperspectral data Tropical Agriculturist, 158, 41-60.
35. Kokaly, R.F. and Clark, R.N. (1999). Spectroscopic determination of leaf biochemistry
usingband-depth analysis of absorption features and stepwise multiple linear regression.
Remote Sensing of Environment, 67, 267-287.
36. Kruse, F.A., Lefkoff, A.B., Boardman, J.B., Heidebrecht, K.B., Shapiro, A.T., Barloon,
P.J. and Goetz, A.F.H. (1993). The Spectral Image Processing System (SIPS) Interactive Visualization and Analysis of Imaging spectrometer Data. Remote Sensing of
the Environment. 44, 145 163.
37. Kruse, F.A., Lefkoff, A.B. and Boardman, J.W. (1993). The Spectral Image Processing
System (SIPS) - Interactive visualization and analysis of imaging spectrometer data.
Remote Sensing of Environment. 44, 145-163.
38. Liang, S. and Strahler, A.H. (1993). Calculation of the angular radiance distribution for
a coupled atmosphere and canopy. IEEE Transactions on Geoscience and Remote
Sensing, 31(2), 491-502.
39. Lillesand, Thomas, M. and Kiefer, R.W. (1987). Remote Sensing and Image
Interpretation, 2nd ed., John Wiley & Sons, New York.
40. Lucas, N.S. and Curran, P.J. (1999). Forest ecosystem simulation modeling: The role of
remote sensing. Progress in Physical Geography. 23(3), 391-423.
41. Myneni,R.B. and Ross, J. (1991). Photon-Vegetation interactions: applications in
optical remote sensing and plant ecology, Springer-Verlag, New York. 565 .
42. Nilson, T. and Kuusk, A. (1989). A reflectance model for the homogeneous plant
canopy and its inversion. Remote Sensing of Environment. 27, 157-167.
II.127

43. Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling, W.T. (1986). Numerical
Recipes. Cambridge University Press, Cambridge. 994.
44. Ranjan, R., Sahoo, R.N., Tomar, R.K,, Nagarajan, S., Gupta, V.K. and Sharma, A.R.
(2008). Characterization of different crop residues and their differential affect on wheat
crop growth with hyper-spectral remote sensing. National Symposium on Advances in
remote sensing technology with special emphasis on microwave remote sensing (ISRS
symposium 2008), Dec. 18-20, Nirma University Campus, Ahmedabad.
45. Sabins, Floyd, F., Jr., (1987). Remote Sensing Principles and Interpretations, 2nd ed.,
W. H. Freeman and Company, New York.
46. Sahoo, R.N. (2008) Potential Application of Hyperspectral Remote Sensing in
Agriculture In National Seminar on Hyperspectral Remote Sensing and Spectral
Signature Database Management System Hyperspec 2008 Feb 14-15, 2008,
Annamalai University, Chidambaram, Tamilnadu.
47. Sahoo, R.N., Bhavanarayana, M., Panda, B.C., Arika, C.N. and Kaur, R. (2005). Total
information content as an index of soil moisture J. Indian Soc. Remote Sensing. 33(1),
17-24.
48. Sahoo, R.N., Tomar, R.K., Pandey, S., Chandana, P., Gupta, R.K., Sharama, R.K.,
Singh, S., Sehgal, V.K., Chakraborty, D. and Kalra, N. (2006). Remote sensing based
growth monitoring of RCT adopted wheat in rice based cropping system in TransGangetic Plain. Presented in 2nd International Rice Congress-Science, Technology and
Trade for Peace and Prosperity 26th International Rice Research Conference, Oct 913, New Delhi: 495.
49. Salisbury, J., Milton, N.M. and Walsh, P.A. (1987). Significance of non-isotropic
scattering from vegetation for geobotanical remote sensing. International Journal of
Remote Sensing, 8, 997-1009.
50. Santra, P., Sahoo, R.N., Das, B.S. and Gupta, V.K. (2009). Estimation of Soil Hydraulic
Properties using Spectral Reflectance in Visible and NearInfrared Region
Geoderma,152, 338-349.
51. Schmidt and Skidmore (2002). Spectral discrimination of vegetation types in a coastal
wetland.Sciences. In: Rencz, A. N. (Ed.): Remote sensing for the Earth sciences:
Manual of remote Sensing of Environment. 67, 267-287.
52. Singh, D.K., Sahoo, R.N., Ahmed, N., and Gupta, V.K. (2008). Quantitative
Assessment of Soil Parameters from Hyperspectral Remote Sensing Data, National
Symposium on Advances in remote sensing technology with special emphasis on
microwave remote sensing (ISRS symposium 2008), Dec. 18-20, Nirma University
Campus, Ahmedabad.
53. Singh, D.K., Ahmed, N., Sahoo, R.N. and Gupta, V.K. ( 2008). Characterization of
some soil orders in relation to hyper-specteral reflectance data, National Symposium on
Advances in remote sensing technology with special emphasis on microwave remote
sensing (ISRS symposium 2008), Dec. 18-20, Nirma University Campus, Ahmedabad.
54. Singh, D.K., Ahmed, N., Sahoo, R.N. and Gupta, V.K. (2009). Determination of Some
Soil Properties using Hyper-Spectral Reflectance Data, Fourth World Congress on
Conservation Agriculture, Feb 4-7, New Delhi, India.
55. Toselli, F. (Ed.), (1989). Applications of Remote Sensing to Agrometeorology, Kluwer
Academic Publishers, Dordrecht, The Netherlands.

II.128

56. Tripathi R., Sahoo, R.N., Sehgal, V.K., Gupta ,V.K., Tomar, R.K. and Chopra, U.K.
(2009). Estimation of biophysical parameters of wheat from MODIS reflectance
through Radiative transfer modeling In ISRS Annual Convention and National
Symposium on Advances in Geo-spatial Technologies with special emphasis on
Sustainable Rainfed Agriculture by Indian Soc. Of Remote Sensing, Nagpur, Sept 1719, 2009.
57. Tripathi, R., Sahoo, R.N., Sehgal, V.K., Tomar, R.K., Chakraborty, D., and Nagarajan
S. (2012). Inversion of PROSAIL Model for Retrieval of Plant Biophysical Parameters.
J. Indian Soc. Remote Sens, 40(1), 19 28.
58. Tripathi, R., Sahoo, R.N., Sehgal, V.K., Tomar, R.K., Pandey, S., Chakraborty, D. and
Kalra, N. ( 2006). Retrieval of plant biophysical parameters through inversion of
PROSAIL model. Proc. of SPIE Vol. 6411, 64110F-1 0277-786X/06/$15. doi:
10.1117/12.697954.
59. Tucker, C.J. ( 1980). Remote sensing of leaf water content in the near infrared. Remote
Sensing of Environment, 10, 23 32.
60. Ustin, S. L., Smith, M. O., Jacquemoud, S. (1999). Geobotany: Vegetation mapping for
Earth Sciences. In: Rencz, A.N. (Ed.): Remote sensing for the Earth sciences: Manual of
remote sensing, 3rd ed., Vol. 3. New York, Chichester, Weinheim. John Wiley &
Sons.189-248.
61. Verhoef, W. and Bunnik, N.J.J.(1981). Influence of crop geometry on multispectral
reflectance determined by the use of canopy reflectance models. Proc.Intl.Colloq.
Signatures Remotely Sens. Objects, Augnon, France, 273-290.
62. Vermote, E.F. and Vermeulen, A.(1999). Atmospheric correction algorithm: Spectral
reflectances (MOD09). Algorithm Theoretical Background Document available online
at http://modarch.gsfc.nasa.gov/data/ ATBD/atbd_mod08.pdf (accessed 23 Feb. 2008)
63. Zarco, Tejada., P.J., Rueda, C.A. And Ustin, S.L.(2003). Water content estimation in
vegetation with MODIS reflectance data and model inversion methods. Remote Sensing
of Environment, 85(1), 109-124.

II.129

You might also like