E-Learning Material: Hyperspectral Remote Sensing

6 Year Integrated M. Tech.
Geotechnology and
Geoinformatics
e-Learning Material
PAPER CODE: MTIGT0901
HYPERSPECTRAL REMOTE SENSING
Dr. J. SARAVANAVEL
Assistant Professor
Centre for Remote Sensing
Bharathidasan University
Tiruchirappalli- 620023
Email: drsaraj@gmail.com
e-learning Material – Hyperspectral Remote Sensing

Dr.J.Saravanavel, Assistant Professor, Centre for Remote Sensing, Bharathidasan University. drsaraj@gmail.com
1
MTIGT0901: HYPERSPECTRAL REMOTE SENSING IN GEOTECHNOLOGY
4 Credits
1. Introduction: Understanding hypersectra, Multispectral vs hyperspectral remote

sensing, Multispectral and Hyperspectral Image Data, Spectral Signatures in the Visible,
Near Infrared and Shortwave Infrared regions of EMR, Hyperspectral Issues.
12 Hrs.
2. Hyperspectral Imaging Devices: Principles of operation, specifications of various sensors -
Spectrographic imagers - hyperspectral sensors, Design tradeoffs. Data formats and systems,
AVIRIS, CASI, MODIS, Hyperion. 12 Hrs.
3. Preprocessing of Hyperspectral Data: Data Cube, Hyperspectral Profiles, Data Redundancy.
Problems with Dimensionality, Principal Component, Minimum Noise Fraction (MNF),
Atmospheric Correction, Atmospheric Correction
Measures, Empirical Line Calibration, Empirical Flat Field Optimized, Reflectance
Transformation (EFFORT), Continuum Removal, Spectral Feature Fitting.
12 Hrs.
4. Hyperspectral Data Analysis: techniques for analysis of hyperspectral data, Derivative
spectral analysis, spectral unmixing, MTMF, case studies.
12 Hrs.
5. Applications: Applications of Hyperspectral lmage Analysis Mineral exploration, Forestry,
soil mapping, water quality studies, quantification of biophysical parameters.
16 Hrs.
References:
1. American Society of Photogrammetry, Manual of Remote Sensing (2nd Edition), ASP

Falls Church, Virginia, 1983.
2. Lillisand, T.M. and Kiefer, P.W, Remote Sensing and Image interpretation, John Wiley
& Sons, New York. 1986
3. Hord R.Michel, Remote Sensing Methods and Application, John Wiley and Sons. 1986.
4. Alexay Bunkin & Konstantin Volia.K, - Laser Remote Sensing of the Ocean Methods &
Publications. John Wiley & Sons, New York, 2001.

Dr.J.Saravanavel, Assistant Professor, Centre for Remote Sensing, Bharathidasan University. drsaraj@gmail.com
2
Hyperspectral Remote Sensing
e – Learning Material: Unit-1
 Since the beginning of time, light was thought to be pure and

singular.
 In 1666, Isaac Newton, separate the light into the color spectrum
using prism
 For longtime, thought that the light to be a bullet-like particle

that traveled in a straight beam
 The scientist like Christian Huygens and Thomas Young prove that
the motion of light followed wave patterns (Young, 1804).

Dr.J.Saravanavel, Assistant Professor, Centre for Remote Sensing 3
Bharathidasan University. drsaraj@gmail.com
 At about the same time in the 1800s, William Herschel observed
that the different Colors actually varied in temperature. Blues
and greens were cooler, while reds were warmer. In fact, he
further discovered that the energy present outside the red
spectrum was warmer yet, and this region became known as
infrared.
 This was the first documented reference that the visible light
spectrum was related to frequency and wavelength.
 These discoveries of the interrelations among color, frequency,

and wavelength laid the framework for hyperspectral remote
sensing because these fundamental principles can be used to
characterize the reflection of light against objects.
Definitions
The term “multi” is derived from the Latin word for
“many” and “hyper” is the Greek word for “over,”
“above,” or an “exaggerated amount.” These,
combined with “spectral,” which relates to colors,
are combined to form “multispectral” and
“hyperspectral,” which figuratively mean “many
colors.”
Differs from conventional remote sensing in that it

covers many narrowly defined spectral channels,
where as, conventional remote sensing looks at
several broadly defined spectral regions.

Hyperspectral remote sensing is the science of acquiring digital
imagery of earth materials in many narrow contiguous spectral
bands.
The simultaneous acquisition of images of the same area in

many (usually 100 or more), narrow, contiguous, spectralbands.
The preferred term is “imaging spectroscopy”.
Hyperspectral sensing (imaging spectrometry) takes remote

sensing from the era of "picturing" into the era of "measurement".
Ultraspectral ("beyond hyperspectral") is used to describe

datasets containing thousands of bands.
Although no such sensor has been designed with this

complexity, it is regarded as the future of imaging
spectroscopy.
Photon
A photon is a discrete particle of electromagnetic energy having
no mass, no electric charge, and an indefinite life. The existence
of photons was first based on the interpretation of experimental
results and presented in a scientific paper by Albert Einstein in
1905.
The energy E of any photon is related to its frequency as follows,

where h is Planck’s constant (6.626068 × 10–34 m2 kg/s) and υ is
the frequency:
E = hυ
Speed of light is ‘c’ and is related to the frequency ‘v’ and the
wavelength ‘λ’ in any part of the electromagnetic spectrum
c = vλ

Electromagnetic Spectrum
Each photon of the electromagnetic spectrum has a wavelength
determined by its energy level. Light and other forms of
electromagnetic radiation commonly are described in terms of
their wavelengths.
Electromagnetic spectral regions (Sabines, 1987)

Electromagnetic Energy
Atmospheric Absorption
Atmospheric Transmittance: Windows for Remote Sensing

Any effort to measure the spectral properties of a material
through a planetary atmosphere, must consider where the
atmosphere absorbs.
The drop toward the ultraviolet is due to scattering and strong

ozone absorption at wavelengths short of 0.35 µm. Ozone also
displays an absorption at 9.6 µm. Oxygen absorbs at 0.76 µm in
a narrow feature. CO2 absorbs at 2.01, 2.06, and a weak doublet
near 1.6 µm. Water causes most of the rest of the absorption
throughout the spectrum
However, these spectral regions can be used in the laboratory

where the atmospheric path lengths are thousands of times
smaller, or when measuring spectra of other planets from
orbiting spacecraft

Emission and Reflection
Photons can be absorbed, reflected, or transmitted. In the realm of
thermodynamics, radiated heat creates photons.
Radiation heat transfer is the exchange of thermal radiation energy between

two or more bodies. Thermal radiation is defined as electromagnetic radiation
in the wavelength range of 0.1 to 100 microns.
Radiation heat transfer must account for both incoming and outgoing thermal
radiation and can be expressed as:
l = εreflected + εabsorbed + εtransmitted

Interaction of energy and objects
Emitted Energy
MW-LWIR
Reflected Energy
Incident Energy V-MWIR
Absorbed Energy
Transmitted Energy
Energy Balance Equation: EI () = ER () + EA() + ET()
Because most solid bodies are opaque to thermal radiation,

transmission can be ignored.
l = εreflected + εabsorbed
To account for emissive radiation, a comparison is made to a
perfect blackbody, which is a theoretical object that absorbs
100% of the incident radiation, reflects none, and appears
perfectly black.
The ratio of the actual emissive radiation E to the emissive

power of a blackbody is defined as the surface emissivity ε.
ε = E/E blackbody

Reflectance is the percentage of incident light that is reflected by
a material.
Reflectance is the ratio of reflected energy to incident energy.
Varies with wavelength

Function of the molecular properties of the material
In climatology and remote sensing, reflectivity is commonly

referred to as “albedo,” the Latin term for white
Reflectivity R can be expressed as:
R= l–ε
Reflected vs. Emitted Energy

MWIR
10 4
radiant exitance (W-m-2-um-1)
1000
Earth
Irradiance (W-m-2-um-1)
Reflectance
100
(100%) Earth
Emission
(100%)
10
Assumes no
atmosphere
1
0.1 .4 .7 1 3 5 7 10
Wavelength (µm)

Emissive Energy
Basic Concepts
• Blackbody – A theoretical material that absorbs and radiates 100% of the energy i ncident
upon i t. BB curve is a function of temperature and wavelength.
• Planck’s Law – gi ves shape of blackbody curve a t a specific temperature.
• Wien’s Displacement Law – determines wavelength of peak emittance.
6000K Peak
Spectral Radiant Emittance
Sun Emittance
373K
1500K Boiling
Hot Coals Water
3000K
Light Bulb 800K
500K
250K
300K
Am bient
W avelength (µm) 0.2 0.4 0.7 1 2 3 5 8 10 30
Reflected Energy
• The manner in which a material reflects energy is primarily a
function of the optical properties and surface roughness of the
feature.
• Most objects are diffuse reflectors
Angle of Incidence = Angle of Reflectance

Energy
Scattered in
All Directions
Smooth Rough
Surface Surface
Specular Diffuse (M icroscopic)
Reflectance Reflectance

Emissivity and reflectivity are two of the fundamental physics
principles that govern hyperpsectral remote sensing.
The portion of the electromagnetic spectrum sought for

exploitation from a hyperspectral sensor—visible, near infrared,
shortwave infrared, midwave infrared, or longwave infrared
Principles of Spectral Radiometry

Radiometry is the physical measurement of electromagnetic
radiation within the ultraviolet, visible, and infrared
wavelengths.
A radiometer is a device used to measure the radiant flux or

power in electromagnetic radiation.
The most important characteristics of a radiometer are spectral

range (wavelengths measured), spectral sensitivity (sensitivity
vs. wavelengths measured), field of view (18 degrees or limited
to a certain narrow field), and directional response

Radiance, Irradiance, and Flux
Radiance and spectral radiance are radiometric measures that
describe the amount of light that passes through or is emitted
from a particular area and falls within a given solid angle in a
specified direction
These measures are used to characterize both emission from

diffuse sources and reflection from diffuse surfaces. The SI unit of
radiance is watts per steradian per square meter (W·sr–1·m–2).
Radiance characterizes total emission or reflection, while spectral

radiance characterizes the light at a single wavelength or
frequency.
The radiance is equal to the sum (or integral) of all the spectral
radiances from a surface.

The SI units for spectral radiance are W·sr–1·m–3 when measured
per unit wavelength, and W·sr–1·m–2·Hz–1 when measured per unit
frequency interval.
Radiance is useful because it indicates how much of the power
emitted by an emitting or reflecting surface will be received by an
optical system looking at the surface from some angle of view
“Irradiance,” “radiant emittance,” and “radiant existence” are
radiometry terms for the power of electromagnetic radiation at a
surface per unit area.
The term “irradiance” is used when the electromagnetic radiation

is incident on the surface.
The SI units for all of these quantities are watts per square meter
(W/m2). These quantities are sometimes called “intensity,”
All of these quantities characterize the total amount of radiation

present, at all frequencies. Each frequency is also commonly
considered in the spectrum separately.
When this is done for a radiation incident on a surface, it is called

“spectral irradiance” and has SI units W/m3, or Commonly W·m–
2 ·nm–1 .
Radiant flux or radiant power is the measure of the total power

of electromagnetic radiation (including visible light). The power
can be the total emitted from a source or the total landing on a
particular surface.

Radiance vs. Reflectance
Radiance is the variable directly measured by remote sensing
instruments.
Radiance is the amount of light the instrument detects from the object
being observed.
When looking through an atmosphere, some light scattered by the
atmosphere will be seen by the instrument and included in the observed
radiance of the target.
An atmosphere will also absorb light, which will decrease the observed
radiance.
Reflectance is the ratio of the amount of light leaving a target to the
amount of light striking the target. If all of the light leaving the target is
intercepted for the measurement of reflectance, the result is called
“hemispherical reflectance.”
Radiance, on the other hand, depends on the illumination (both its
intensity and direction), the orientation and position of the target, and the
path of the light through the atmosphere.
NASA AVIRIS Cuprite, NV, HSI Data, (1995)

Spectral comparison between hyperspectral and broad-
band data

Multi Spectral vs Hyperspectral
Multispectral - Hyperspectral
Signature Comparison
Multispectral Hyperspectral
Resampled to Landsat TM7 Bands

Levels of Spectral Information
Multi Spectral vs Hyperspectral
High Spectral Resolution
Qua ntification: Determines the a bundance of
ma terials.
Cha ra cterization: Determines va riability of i dentified
ma terial (e.g. wet/dry s and, s oil particle size effects).
Hyperspectral Identification: Determines the unique i dentity of the
(100’s of bands) foregoing generic categories (i.e. material
i dentification).
Di s crimination: Determines generic categories of the
foregoing cl asses.
Multispectral
(10’s of bands) Cl a ssification: Separates materials i nto spectrally
s i milar groups.
Detection: Determines the presence of materials,
Panchromatic objects, a ctivities, or events.
Low Spectral Resolution

Multispectral vs. hyperspectral remote
sensing
Multispectral Hyperspectral
• separated spectral bands • no spectral gaps
• wider bandwidths • narrow bandwidths (10nm)
• coarse representation of the • complete representation of the
spectral signature spectral signature
• not able to discern small • ability to detect subtle spectral
differences between reflectance features
spectra • large data volumes
• smaller data volumes • radiometric and spectral
• fewer problems with calibration calibration are time-consuming
• multi-decadal history of • currently no functioning
continuous image acquisition sensors in orbit
33
Limitations and Issues of Hyperspectral Data
Very sensitive to noise
Difficult to interpret the spectral signatures of an “impure” pixel
It need for calibration
• Because of the high spectral resolution of hyperspectral

imaging, fine atmospheric absorption features will be detected,
which may be confused with the ground material being imaged

Mixed Pixel Problem
 Presence of small or sub pixel targets
 Presence of boundaries of discrete

parcels
 Gradual transition between land

cover classes
 Contribution of areas outside the area

represented by a pixel
The spectral reflectance of a mixed pixel is generally the

weighted average of the spectral response of the classes
within it
Data volume
• storage and transmission of hyperspectral data are one of the
issues (AVIRIS=40*TM)
Redundancy
• overlap of information content over several bands (not all data
is needed to characterize a pixel properly)

Hyperspectral data characteristics
•Hyperspectral datasets generally contain at least 16
contiguous bands of high spectral resolution over a
region of the electromagnetic spectrum.
•Advancements in solid-state technology in recent years

have made possible collection of multispectral data in as
many as 200 or more (~288) spectral bands.
• Most are able to collect images starting at about 400

nm which is the edge of the blue visible part of the
spectrum.
Typically these systems can measure energy to

1100 or even 2500 nm.
Hyperspectral sensors or imaging spectrometers measure

earth materials and produce complete spectral signatures
with no wavelength omissions.
Such instruments are flown aboard space and air-based

platforms.

To understand the Hyper spectral Remote Sensing, it is necessary
to have a basic knowledge of spectroscopy in the 0.4 to 2.5µm
region covered by the sensors.
Spectroscopy is the study of light as a function of wavelength

that has been emitted, reflected or scattered from a solid,
liquid, or gas
Absorption and Scattering. As photons enter a mineral, some are

reflected from grain surfaces, some pass through the grain, and
some are absorbed.
Those photons that are reflected from grain surfaces or refracted

through a particle are said to be scattered. Scattered photons
may encounter another grain or be scattered away from the
surface so they may be detected and measured.
Photons may also originate from a surface, a process called

emission. All natural surfaces emit photons when they are above
absolute zero

Photons are absorbed in minerals by several processes. The
variety of absorption processes and their wavelength
dependence allows us to derive information about the
chemistry of a mineral from its reflected or emitted light.
The human eye is a crude reflectance spectrometer: we can look

at a surface and see color. Our eyes and brain are processing the
wavelength-dependent scattering of visible-light photons to
reveal something about what we are observing, like the red
color of hematite or the green color of olivine.
A modern spectrometer, however, can measure finer details

over a broader wavelength range and with greater precision.
Thus, a spectrometer can measure absorptions due to more
processes than can be seen with the eye.
Spectroscopy Terms There are 4 general parameters that describe

the capability of a spectrometer:
1) spectral range,
2) spectral bandwidth,
3) spectral sampling, and
4) signal-to-noise ratio (S/N).
There are general spectral ranges that are in common use, each to
first order controlled by detector technology: a) ultraviolet (UV):
0.001 to 0.4 µm, b) visible: 0.4 to 0.7 µm, c) near-infrared (NIR): 0.7
to 3.0 µm, d) the mid-infrared (MIR): 3.0 to 30 µm, and d) the far
infrared (FIR): 30 µm to 1 mm
Spectral bandwidth is the width of an individual spectral channel in
the spectrometer. The narrower the spectral bandwidth, the
narrower the absorption feature the spectrometer will accurately
measure, if enough adjacent spectral samples are obtained

Some systems have a few broad channels, not contiguously spaced and, thus,
are not considered spectrometers. Examples include the Landsat Thematic
Mapper (TM) system and the MODerate Resolution Imaging
Spectroradiometer (MODIS), which can't resolve narrow absorption features
Like the NASA JPL Airborne Visual and Infra-Red Imaging

Spectrometer (AVIRIS) system, VIMS, NIMS, HYDICE have many
narrow bandwidths, contiguously spaced. Bandwidths and
sampling greater than 25 nm rapidly lose the ability to resolve
important mineral absorption features

The shape of the bandpass profile is also important. Ideally
each spectrometer channel rejects all light except that from
within a given narrow wavelength range, but occasionally, due
to optical effects, light may leak in from out of the bandpass
(e.g. scattering within the optical system, or inadequate
blocking filters).
The most common bandpass in spectrometers is a Gaussian

profile. While specific spectrometer designs may have well-
defined theoretical bandpass profiles, aberrations in the optical
system usually smears the profile closer to a Gaussian shape.
The width of the bandpass is usually defined as the width in

wavelength at the 50% response level of the function, as shown
in Fig, called the Full Width at Half Maximum (FWHM).
A Gaussian profile with a Full Width at Half Maximum (FWHM) of 10 nm is

shown. This profile is typical of spectrometers such as AVIRIS which has 224
such profiles spaced at about 10 nm

Spectral sampling is the distance in wavelength between the
spectral bandpass profiles for each channel in the spectrometer
as a function of wavelength. Spectral sampling is often confused
with bandpass, with the two lumped together and called
resolution
The signal-to-noise ratio (S/N) required to solve a particular

problem will depend on the strength of the spectral features
under study.
The S/N is dependent on the detector sensitivity, the spectral

bandwidth, and intensity of the light reflected or emitted from
the surface being measured.
A few spectral features are quite strong and a signal to noise of

only about 10 will be adequate to identify them, while others are
weak, and a S/N of several hundred (and higher) are often
needed (Swayze et al., 1997).

Imaging Spectroscopy
Today, spectrometers are in use in the laboratory, in the field, in aircraft
(looking both down at the Earth, and up into space), and on satellites.
Reflectance and emittance spectroscopy of natural surfaces are sensitive
to specific chemical bonds in materials, whether solid, liquid or gas
Spectroscopy's historical disadvantage is that it is too sensitive to small
changes in the chemistry and/or structure of a material. The variations in
material composition often causes shifts in the position and shape of
absorption bands in the spectrum.
Thus, with the vast variety of chemistry typically encountered in the real
world, spectral signatures can be quite complex and sometimes
unintelligible.
However, that is now changing with increased knowledge of the natural
variation in spectral features and the causes of the shifts. As a result, the
previous disadvantage is turning into a huge advantage, allowing us to
probe ever more detail about the chemistry of our natural environment.
With the advances in computer and detector technology, the new

field of imaging spectroscopy is developing.
Imaging spectroscopy 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.
By analyzing the spectral features, and thus specific chemical

bonds in materials, one can map where those bonds occur, and
thus map materials. Such mapping is best done, in this author's
opinion, by spectral feature analysis.

Imaging spectroscopy has many names in the remote sensing
community, including imaging spectrometry, hyperspectral, and
ultraspectral imaging.
Imaging spectroscopy for remote sensing involves the acquisition of

image data in many contiguous spectral bands with an ultimate goal
of producing lboratory quality reflectance spectra for each pixel in
an image (Goetz,1992)
Spectroscopy is the study of electromagnetic radiation.

Spectrometry is derived from spectro-photometry, the measure of
photons as a function of wavelength.

THE REFLECTION AND ABSORPTION PROCESSES
When a stream of photons encounter a medium with a change in the index of
refraction some are reflected and some are refracted into the medium
All materials have a complex index of refraction

m = n - jK
where m is the complex index of refraction, n is the real part of the index, j = (-
1)1/2 , and K is the imaginary part of the index of refraction, sometimes called
extinction coefficient
When photons enter an absorbing medium, they are absorbed according to

Beers Law:
I = Io e -kx, (eqn 1b)
where I is the observed intensity, Io is the original light intensity, k is an
absorption coefficient and x is the distance traveled through the medium
As distance increases through a medium, more photons will be intercepted

The rate of interception of photons is higher with shorter wave length
The absorption coefficient is related to the complex index of

refraction by the equation:
k = 4  K/ , (eqn 1c)
where ‘ ’ is the wavelength of light.
The reflection of light, R, normally incident onto a plane surface is

described by the Fresnel equation
R = [(n - 1)2 + K2 ] / [(n + 1)2 + K2 ]. (eqn 1d)

At angles other than normal, the reflectance is a complex
trigonometric function involving the polarization direction of the
incident beam .
R = Reflection of light normally incident onto plane surface

The reflection from quartz grains as measured on a laboratory spectrometer
The absorption coefficient (k = 4  K/) determine how far

into a material light of a particular wave length can
penetrate before it is absorbed . In a material with low a low
absorption coefficient, light is only poorly absorbed
The absorption coefficient as a function of wavelength is a

fundamental parameter describing the interaction of photons with
a material.
So is the index of refraction, but it generally varies less than the

absorption coefficient as a function of wavelength, especially at
visible and near-infrared wavelengths
At fundamental absorption bands, both n and K vary strongly with

wavelength, though K still varies over more orders of magnitude
than n.
The complex interaction of light with matter involves reflection and

refraction from index of refraction boundaries, a process we call
scattering, and absorption by the medium as light passes through the
medium. The amount of scattering versus absorption controls the
amount of photons we receive from a surface.

The complex index of refraction important properties of materials.
As one moves to longer wavelengths, the index of refraction
decreases to a minimum just before a sharp rise (e.g. at 8.5 and
12.6 µm). The minimum is often near or even below n = 1. The
wavelength where n = 1 is called the Christensen frequency and
usually results in a minimum in reflected light because of the small
(to zero) difference in the index of refraction compared to the
surrounding medium (e.g. air or vacuum).
The location of the observed reflectance minimum is also

controlled by the extinction coefficient according to equation 1d.
This maximum is called the restrahlen band: the location of

fundamental vibrational stretching modes in the near and mid-
infrared. The combination of n and K at these wavelengths often
results in high reflectance.

So the Christensen frequency
and Restrahlen peak are the
characteristic features of
individual group of minerals and
hence their fundamental
absorption wave length
The complex interaction of light

with matter involves reflection
and refraction from index of
refraction boundaries, a process
we call scattering, and
absorption by the medium as
light passes through the
medium. The amount of
scattering versus absorption
controls the amount of photons
we receive from a surface.
CAUSES OF ABSORPTION
Isolated atoms or ions have discrete energy states. A change of
the energy state is referred to as transition. If a photon of a
specific wavelength is absorbed or emitted, the energy state is
changed.
Absorption results in a higher energy state and emission in a

lower energy state.
Since an absorbed photon of a specific wavelength is usually not

emitted at the same wavelength, absorption bands can be
generated.
Absorption bands within the visible and reflected infrared light

are related to electronic and vibrational processes within the
crystal lattice.

Electronic transition processes require more energy; hence they
occur mainly at shorter wavelengths within the VIS and NIR
spectrum, and give rise to broad absorption features.
Vibrational processes take place within the infrared spectra (NIR,

SWIR and thermal) and result typically in small and sharp
absorption bands.
The position and shape (depth, width and asymmetry) of an

absorption feature are controlled primarily by the kind of absorbing
ion or molecule and its position within the crystal lattice (e.g.
octahedral, tetrahedral site). But also the kind of chemical bond
and the elements involved in the bond influence the position and
shape of the absorption feature.
Electronic processes
In minerals, electronic transition processes yielding absorption features
are related to crystal field effects, conduction bands, charge transfer
transition and colour centres. Which processes take effect depends on
the chemical bond and the involved elements.
Crystal field effects: The most common electronic process revealed

in the spectra of minerals i s due to unfilled electron shells of
transition elements (Ni, Cr, Co, Fe, etc.). Iron is the most common
transition element in minerals.
For all transition elements, d orbitals have identical energies in an

isolated ion, but the energy levels split when the atom is located in a
crystal field (e.g. see Burns, 1970, 1993).
This splitting of the orbital energy state s enables an electron to be

moved from a lower level into a higher one by absorption of a photon
having an energy matching the energy difference between the states.

The energy levels are determined by the valence state of the atom
(e.g. Fe2+, Fe3+), its coordination number, and the symmetry of the
site it occupies.
The levels are also influenced by the type of ligands formed, the
extent of distortion of the site, and the value of the metal-ligand
interatomic distance (e.g. Burns, 1993).
The crystal field varies with crystal structure from mineral to

mineral, thus the amount of splitting varies and the same ion (like
Fe2+) produces obviously different absorptions, making specific
mineral identification possible from spectroscopy
Reflectance spectra of two olivines showing the change in band position and
shape with composition. The 1-µm absorption band is due to a crystal field
absorption of Fe 2+.
The Fo 29 sample has an FeO content of 53.65%, while the Fo 91 sample has an
FeO content of 7.93%.
The mean grain size is 30 and 25 µm respectively. The 1-µm band position varies
from about 1.08 µm at Fo 10 to 1.05 µm at Fo 90 (King and Ridley, 1987).

Reflectance spectra of
two pyroxenes showing
the change in Fe2+ -
absorption band
position and shape with
composition (from Clark
et al., 1993b).
Diopside is CaMgSi2O6, but some Fe2+ substitutes for Mg.
Bronziteis (Mg,Fe)SiO3 with mostly Mg.
The 1-µm versus the 2-µm band position of a pyroxene describes the
pyroxene composition
Charge Transfer Absorptions

Absorption bands can also be caused by charge transfers, or inter-
element transitions where the absorption of a photon causes an
electron to move between ions or between ions and ligands.
The transition can also occur between the same metal in different
valence states, such as between Fe2+ and Fe3+.
In general, absorption bands caused by charge transfers are

diagnostic of mineralogy.
Their strengths are typically hundreds to thousands of times

stronger than crystal field transitions.
The band centers usually occur in the ultraviolet with the wings of
the absorption extending into the visible. Charge transfer
absorptions are the main cause of the red color of iron oxides and
hydroxides.

Reflectance spectra of the i ron oxi de hema ti te (Fe2O3) and iron hydroxide goethite
(FeOOH, from Cla rk et al., 1993b). The intense charge-tra nsfer band in the UV (< 0.4 µm) is
"satura ted" in reflectance, so onl y fi rst surfa ce (specula r) reflection is seen in these spectra.
The 0.9-µm and 0.86-µm abs orption fea tures a re due to Laporte-forbidden transitions (e.g.
Morris et al , 1985; Sherman, 1990 a nd references therein). The absorption a t 2.7-3 µm is
due to tra ce water in the samples ., and in the case of goethite, the OH. The goethite
s pectrum is offset upward 0.2 units.
Conduction Bands. In some minerals, there are two energy levels

in which electrons may reside: a higher level called the
"conduction band," where electrons move freely throughout the
lattice, and a lower energy region called the "valence band,"
where electrons are attached to individual atoms.
The difference between the energy levels is called the band gap.
The band gap is typically small or non-existent in metals, and very
large in dielectrics.
In semiconductors, the band gap corresponds to the energy of

visible to near-infrared wavelength photons and the spectrum in
these cases is approximately a step function.
The yellow color of sulfur is caused by such a band gap. The

minerals cinnabar (HgS) and Sulfur (S) have spectra showing the
band gap in the visible

Absorption occurs when photons have sufficient energy to
excite electrons across the forbidden band. Many sulphides
show absorption features related to conduction bands.
Reflectance spectra of Sulfur, S, (top) and cinnabar, HgS, (bottom) showing

conduction bands in the visible (from Clark et al., 1993b).

Color Centers. A few minerals show color due to absorption by
"color centers." A color center is caused by irradiation (e.g. by
solar UV radiation) of an imperfect crystal.
Crystals in nature have lattice defects that disturb the periodicity

of the crystal. For example, defects might be caused by impurities.
These defects can produce discrete energy levels and electrons can
become bound to them. The movement of an electron into the
defect requires photon energy. The yellow, purple and blue colors
of fluorite are caused by color centers. See Hunt (1977) and
references therein for more details.

Vibrational Processes
When light at the same wavelength as a molecule (or part of a molecule)
strikes the molecule, it causes the molecule (or part of the molecule) to
vibrate. This leads to light absorption. In general these absorptions are
very narrow, although their widths and depths vary. Most of these
absorptions can be detected with hyperspectral sensors.
The bonds in a molecule or crystal lattice are like springs with attached
weights: the whole system can vibrate. The frequency of vibration
depends on the strength of each spring (the bond in a molecule) and
their masses (the mass of each element in a molecule).
For a molecule with N atoms, there are 3N-6 normal modes of vibrations
called fundamentals. Each vibration can also occur at roughly multiples
of the original fundamental frequency. The additional vibrations are
called overtones when they involve multiples of a single fundamental
mode, and combinations when they involve different modes of
vibrations.
Vibrational Processes
Symmetrical Antisymmetrical Scissoring

stretching stretching
Rocking Wagging Twisting

A vibrational absorption will be seen in the
infrared spectrum only if the molecule
responsible shows a dipole moment (it is said to
be infrared active).
Reflectance spectra of calcite,

dolomite, beryl, gypsum, alunite,
rectorite, and jarosite showing
vibrational bands due to OH, CO3
and H2O, from Clark et al., 1990a)
Reflectance spectra of phlogopite, biotite,

pyrophyllite, muscovite, epidote, and illite
showing vibrational bands due to OH and
H2O, from Clark et al., 1990a).

Reflectance spectra of hectorite,
halloysite, kaolinite, chrysotile,
lizardite, and antigorite showing
vibrational bands due to OH
Mid-infrared reflectance spectra of quartz. The strong 9-µm Si-O-Si asymmetric

stretch fundamental is obvious from the reflection maximum.
The O-Si-O bending mode occurs near 25 µm and is the second strongest
absorption.
The absorption between 12 and 13 µm is the Si-O-Si symmetric stretch
fundamental.

Water and Hydroxyl. Water and OH (hydroxyl) produce particularly
diagnostic absorptions in minerals. The water molecule (H2O) has N=3, so
there are 3N-6=3 fundamental vibrations.
In the isolated molecule (vapor phase) they occur at 2.738 µm (v1,

symmetric OH stretch), 6.270 µm (v2, H-O-H bend), and 2.663 µm (v3,
asymmetric OH stretch).
In liquid water, the frequencies shift due to hydrogen bonding: v1=3.106

µm, v2=6.079 µm, and v3=2.903 µm.
The overtones of water are seen in reflectance spectra of H2O-bearing

minerals.
The first overtones of the OH stretches occur at about 1.4 µm and the
combinations of the H-O-H bend with the OH stretches are found near 1.9
µm.

Thus, a mineral whose spectrum has a 1.9-µm absorption band contains
water (e.g. hectorite and halloysite), but a spectrum that has a 1.4-µm
band but no 1.9-µm band indicates that only hydroxyl is present (e.g.
kaolinite has only a small amount of water because of the weak 1.9-µm
absorption but a large amount of OH).
The hydroxyl ion has only one stretching mode and its wavelength position
is dependent on the ion to which it is attached. In spectra of OH-bearing
minerals, the absorption is typically near 2.7 to 2.8 µm, but can occur
anywhere in the range from about 2.67 µm to 3.45 µm (e.g. see Clark et
al., 1990 and references therein).
The OH commonly occurs in multiple crystallographic sites of a specific

mineral and is typically attached to metal ions. Thus, there may be more
than one OH feature. The metal-OH bend occurs near 10 µm (usually
superimposed on the stronger Si-O fundamental in silicates). The
combination metal-OH bend plus OH stretch occurs near 2.2 to 2.3 µm and
is very diagnostic of mineralogy (e.g. see Clark et al., 1990 and references
therein).
Carbonates. Carbonates also show diagnostic vibrational

absorption bands.
The observed absorptions are due to the planar CO3-2 ion. There are
four vibrational modes in the free CO3-2 ion: the symmetric stretch,
v 1: 1063 cm-1 (9.407 µm); the out-of-plane bend, v 2: 879 cm-1 (11.4
µm); the asymmetric stretch, v 3: 1415 cm-1 (7.067 µm); and the in-
plane bend, v 4: 680 cm-1 (14.7 µm)
Combination and overtone bands of the CO3 fundamentals occur in
the near IR. The two strongest are v 1 + 2v 3 at 2.50-2.55µm (4000-
3900 cm-1), and 3v3 at 2.30-2.35 µm (4350-4250 cm-1; e.g. Figure
10a). Three weaker bands occur near 2.12-2.16 µm (v1 + 2v3 + v 4 or
3v 1 + 2v 4; 4720-4630 cm-1), 1.97-2.00 µm (2v1 + 2v 3; 5080-5000 cm-
1 ), and 1.85-1.87 µm (v + 3v ; 5400-5350 cm-1 ; (e.g. Hunt and
1 3
Salisbury, 1971). The band positions in carbonates vary with
composition (Hunt and Salisbury, 1971; Gaffey, 1986, Gaffey et al.,
1993). An example of such a band shift in absorption position from
calcite to dolomite.

SPECTRA OF MISCELLANEOUS MINERALS AND MATERIALS
Organics. Organic materials are found all over the Earth, and in the solar system.
The C-H stretch fundamental occurs near 3.4 µm, the first overtone is near 1.7
µm, and a combination band near 2.3 µm . The combinations near 2.3 µm can
sometimes be confused with OH and carbonate absorptions in minerals
especially at low spectral resolution. Reflectance spectra of
montmorillonite, and
montmorillonite mixed
with super unleaded
gasoline, benzene,
Tra ns mittance toluene, and
s pectra of trichlorethylene.
orga nics a nd Montmorillonite has an
mi xtures absorption feature at 2.2
s howing the µm, whereas the
organics have a CH
compl ex
combination band near
a bs orptions in 2.3 µm. The first
the CH-s tretch overtone of the CH
fundamental stretch can be seen at
s pectral region. 1.7 microns, and the
second overtone near
1.15 µm. From King and
Clark (1989b).
Ices
Just like water in minerals show s
diagnostic absorption bands, ice
(crystalline H2 O) which is formally a
mineral, also shows strong absorption
bands. Spectra of solid H2 O, CO2 , and
CH4 in Figure. The spectral features in
Figure are all due to vibrational
combinations and overtones. Note the
H2 O spectra show broad absorptions
compared to the others. The reason is
that while ice is norm ally a hexagonal
structure, the hydrogen bonds are
orientationally disordered (e.g. Hobbs,
1974), and the disorder broadens the
absorptions.
Reflectance spectra of solid carbon

dioxide, CO2, methane, CH4, and water,
H2O, from Clark et al. (1986).

Vegetation
Spectra of vegetation come in two general forms: green and wet (photosynthetic),
and dry non-photosynthetic but there is a seemingly continuous range between
these two end members.
The spectra of these two forms are compared to a soil spectrum in Figure. Because
all plants are made of the same basic components, their spectra appear generally
similar.
The near-infrared spectra of green vegetation are dominated by liquid water

vibrational absorptions.
The water bands are shifted to slightly shorter wavelengths than in liquid water,
due to hydrogen bonding. The absorption in the visible is due to chlorophyll. The
dry non-photosynthetic vegetation spectrum shows absorptions due to cellulose,
lignin, and nitrogen. Some of these absorptions can be confused with mineral
absorptions, unless a careful spectral analysis is done
Reflectance spectra of photosynthetic (green) vegetation, non-photosynthetic (dry)

vegetation, and a soil. The green vegetation has absorptions short of 1 µm due to
chlorophyll. Those at wavelengths greater than 0.9 µm are dominated by liquid water. The
dry vegetation shows absorptions dominated by cellulose, but also lignin and nitrogen.
These absorptions must also be present in the green vegetation, but can be detected only
weakly in the presence the stronger water bands.

THE SENSITIVITY OF ABSORPTION BANDS TO CRYSTAL STRUCTURE AND CHEMISTRY
In many cases spe ctroscopy is very sensitive to subtle changes in crystal
structure or chemistry. This has resulted in confusion in the past over cause and
effect. More recently, this sensitivity has been recognized as a powerful means
of studying the structure and composition of minerals
Because spectroscopy is sensitive to so many processes, the spectra can be very

complex and there is still much to le arn. However, it is because of this sensitivity
that spectroscopy has great potential as a diagnostic tool. In fact for some
materials, spectroscopy is an excellent tool not only for detecting certain
chemistries, but also at abundance levels unmatched by other tools.
Pyroxenes. The iron bands near 1 and 2 µm shift with pyroxene

composition
The change in Fe2+ -absorption band position and shape with

composition

OH Examples. The sharper OH-related absorption bands
allow smaller band shifts to be measured. These bands can
be so sensitive that it is possible to distinguish between the
isochemical end-members of the Mg-rich serpentine group,
chrysotile, antigorite, and lizardite (King and Clark, 1989a,
and Figure).
Mustard (1992) calibrated changes in the 1.4-µm and 2.3-
µm absorptions in the tremolite-actinolite solid solution
series; example spectra of the 1.4-µm absorptions are
shown in Figure.
High spectral resolution reflectance spectra of the first
overtone of OH in talc, tremolite, actinolite, crysotile,
lizardite, and antigorite. The three sharp absorption bands
in talc, tremolite and actinolite are caused by Mg and Fe
ions associated with the hydroxyls, causing small band
shifts. The Fe:Fe+Mg ratio can be estimated. In chrysotile,
lizardite and antigorite, the absorptions change with small
structural differences even though the composition is
constant
The strong and sharp OH features have proven
particularly diagnostic of clay mineralogy
Al in Muscovite
More recently, subtle shifts have been found in muscovite series with aluminum
composition. As elements substitute for aluminum in the crystal structure, the
crystal becomes slightly distorted relative to no substitutions.
This cause s slight changes in Al-O-H bond lengths and thus shifts absorption
band position. In this case , the shift of the 2.2-µm absorption appears
continuous with composition.
Reflectance spectra of muscovite showing band shifts due to changing aluminum

composition

Reflectance spectroscopy can be used without sample preparation, and it
is non-destructive. This makes mapping of exposed minerals from aircraft
possible, including detailed clay mineralogy (e.g. Clark et al., 1993a).
Visual and near-infrared spectroscopy, on the other hand, is insensitive to

some minerals which do not have absorptions in this wavelength region.
For example, quartz has no diagnostic spectral features in the visible and
near-infrared.
Quartz must be detected at its fundamental Si-O stretching region near 10

µm.
FACTROS AFFECTING SPECTRA

THE SCATTERING PROCESS
Scattering is the process that m akes reflectance spectroscopy possible: photons
enter a surface, are scattered one or more times, and while some are absorbed,
others are scattered from the surface so that we may see and detect them.
Scattering can also be thought of as scrambling information. The information is

made more complex, and because scattering is a non-line ar process, recovery of
quantitative information is more difficult.
Reflectance of a particulate surface, however, is much more complex and the

optical path of photons is a random walk. At each grain the photons encounter,
a certain percentage are absorbed. If the grain is bright, like a quartz grain at
visible wavelengths, most photons are scattered and the random walk process
can go on for hundreds of encounters.
If the grains are dark, like magnetite, the majority of photons will be absorbed
at each encounter and essentially all photons will be absorbed in only a few
encounters

The stronger absorption will penetrate less into the surface , encountering fewer
grains because the photons are absorbed.
At the wavelengths of the weaker absorption, less photons are absorbed w ith
each encounter with a grain, so the random walk process goe s further,
increasing the average photon path length.
The greater path length will result in more absorption, thus strengthening the
weak absorption in a reflectance spectrum.
Mixtures
The real world is a complex mixture of materials, at any scale we view it. In
general, there are 4 types of mixtures
1) Linear Mixture. The materials in the field of view of the detector are optically
separated, so the reflectance at each wavelength is the fraction of each materials.
This is also called areal mixture.
2) Intimate Mixture. An intimate mixture occurs when different materials are in
intimate contact in a scattering surface, such as the mineral grains in a soil or rock.
Depending on the optical properties of e ach component, the resulting signal is a
highly non-linear combination of the end-member spectra.
3) Coatings. Coatings occur when one m aterial coats another. Each coating is a
scattering/transmitting layer whose optical thickness varies with material
properties and wavelength.
4) Mole cular Mixtures. Molecular mixtures occur on a molecular level, such as two
liquids, or a liquid and a solid mixed together. Examples: water adsorbed onto a
mineral; gasoline spilled onto a soil. The close contact of the mixture components
can cause band shifts in the adsorbate, such as the interlayer water in
montmorillonite, or the water in plants.

An example mixture comparison for alunite and jarosite. Note in the intimate
mixture how the jarosite dominates in the 0.4 to 1.3-µm region.
The reason is be cause in an intimate mixture, the darker material dominates
because photons are absorbed when they encounter a dark grain.
In the areal mixture, the brighter material dominates. because the jarosite is
finer grained than the alunite and tends to coat the larger alunite grains.

Grain Size Effects
The amount of light s cattered and absorbed by a grain is dependent on grain size (e.g.
Ha pke, 1993; Cla rk and Roush, 1984). A larger grain has a greater internal pa th where
photons ma y be absorbed a ccordi ng to Beers Law. In a smaller grain there are
proporti onall y more surfa ce reflections compa red to internal photon pa th lengths . The
refl ectance decreases as the grain size increases.
Absorption
band contrast
varies with
particle size but
does not affect
positions of
absorption
features
The near-infrared spectral reflectance of A) a fine grained (~50 µm) water frost, B) medium
grained (~200 µm) frost, C) coarse grained (400-2000 µm) frost and D) an ice block
containing abundant microbubbles. The larger the effective grain size, the greater the mean
photon path that photons travel in the ice, and the deeper the absorptions become. Curve D
is very low in reflectance because of the large path length in ice

The Continuum and Band Depth
Absorptions in a spectrum have two
components: continuum and
individual features.
The continuum is the "background
absorption" onto which other
absorption features are
superimposed (Clark and Roush,
1984). It may be due to the wing of a
larger absorption feature.
For example, in the pyroxene

spectra, the weak feature at 2.3 µm
is due to a trace amount of tremolite
in the sample and the absorption is
superimposed on the broader 2-µm
pyroxene band. The broader
pyroxene absorption is the
continuum to the narrow 2.3-µm
feature.
Spectra are
normalized to a
common reference
using a continuum
formed by defining
high points of the
spectrum (local
maxima) and fitting
straight line segments
between these
points. The
continuum is removed
by dividing it into the
original spectrum.
Source: ENVI Manual
A fitted continuum (bottom) and a continuum-

removed (top) spectrum for the mineral kaolinite

Continuum-Removed Spectral Feature Comparison.
The continuum-removal process isolates spectral features
and puts them on a level. Conti nuum removal was applied
to the s elected absorption features. Continuum removal
norma lizes reflectance spectra in order to allow comparison
of i ndividual a bsorption features from a common baseline
For exa mple, compa re the spectra of calci te (Ca CO3) and
dolomi te (Ca Mg(CO3) 2. If we isola te the spectral features,
remove the continuum, and s cale the ba nd depth (or band
a rea) to be equal, we can see subtle band shi fts and
s ha pes.
Compare a harder case: halloysite and kaolinite. You might

note that halloysite has a different absorption feature at 1.9
µm. However, if you were obtaining the spectrum through the
Earth's atmosphere, you would have virtually no data in that
wavelength region because atmospheric water absorbs too
much of the signal. The diagnostic feature is the 2.2-µm band.
The continua removed 2.2-µm features for halloysite and
kaolinite, where we can see significant differences between
the spectra of the two minerals.
Comparison of kaolinite and halloysite spectral features.

Both mineral spectra have the same band position at 2.2
µm. However, the kaolinite spectrum shows a stronger
feature at 2.16 µm than in the halloysite spectrum

Viewing Geometry
We have seen tremendous variation in the spectral properties of minerals and

materials in general, due to composition, grain size, and mixture types. So far
viewing geometry has not been discussed.
Viewing geometry, including the angle of incidence, angle of reflection, and the
phase angle: the angle between the incident light and observer (the angle of
reflection), all affect the intensity of light received.
Varying the viewing geometry results in change s in shadowing and the

proportions of first surface to multiple scattering (e.g. Hapke , 1993; Nelson,
1986; Mustard and Pieters, 1989), which can affect band depths a sm all
amount except in rare cases (like extreme specular reflection off a mirror or
lake surface). While measuring precise light levels are important for things like
radiation balance studies, they are of lesser importance in spectral analysis

Hyperspectral Sensors
Airborne
1. CASI (Canadian technology)
2. AVIRIS (American NASA technology)
Other airborne:
1. HYDICE
2. DAIS
Spaceborne
1. Hyperion
2. Modis
3. CHRIS
4. MERIS
Ground Based
1. Spectroradiometer

AVIRIS: Airborne Visible Infrared Imaging Spectrometer
Airborne Visible Sensor/Infrared Im aging Spe ctrometer (AVIRIS) are new in
terms of hyperspectral systems attached to planes. The AVIRIS sensor,
developed by NASA/Jet Propulsion Laboratory (JPL), is working in bands of
224, with a spectral resolution of about 10 nm and covering the spectral
range from 0.40 to 2.50 μm. The sensor is a Whiskbroom system that uses a
scanning system for acquiring data on the transverse direction of
advancement.
Four off-axis double-pass Schmidt spectrometers capture light from oreoptics

using optical fiber and send it to four linear panels, one for e ach
spectrometer, which have a strong sensitivity in the range 0.4…0.7 μm, 0.7…
1.2 μm, 1.2…1.8 μm and 1.8…2.5 μm.
AVIRIS sensor takes images from an altitude of 20 km with a spatial

resolution of 20 meters, from a band whose width is of 10.5 kilometers.
Starting with 1998, the sensor is mounted on a Twin Otter aircraft flying at
low altitude, taking pictures with a spatial resolution ranging between 2 and
4 meters
AVIRIS: Airborne Visible Infrared Imaging Spectrometer
 CCD array for spectral bands

 across-track scanning.
 224 spectral bands
 Wavelength range: 0.38 to 2.50 μm.
 Bandwidth: 10 nm
 30° field of view
 614 pixels per scan
 Usually flown on the NASA/ARCER-2
aircraft at an altitude of 20 km
– Swath width 11 km
– Pixel size 20 m x 20 m
Can be flown lower (under clouds)
– Pixel size reduced to ~ 5 m
 12 bit data

HyMAP: the hyperspectral mapper
Typical:
• 128 band
• 400 – 2500 nm.
• 4-5 m. on the ground
• General
– 100 - 200 bands
– bandwidths of 10 – 20 nm
– high signal to noise, > 500:1
– 2 -10 m spatial resolution
– 60 -70 degrees swath width
Integrated Spectronics, Sydney, Australia, has been developing field
spectrometers for several years. Its field portable spectrometer, the
PIMA-II has brought fundamental changes to the way many field and
mine geologists work. With the introduction of the HyMapTM
hyperspectral airborne system, the potential exists to start routine
commercial use of airborne imaging spectrometry. HyMap is the first
commercially available sensor to provide combined high spatial
resolution (2 to 10-m ), high spectral resolution (12 to 16nm), and high
signal-to-noise (>500:1) performance.

CASI Sensor: Compact Airborne Spectrographic Imager
 Manufactured by Itres Research Ltd. of
Calgary
 288 selectable spectral bands ranging from
390-900 nm
 512 spatial pixels across a 35 degree swath
 spatial resolution 1 to 10m depending on
altitude
 radiometric resolution is 12 bit
Other Airborne Sensors

HYDICE
 developed by the US Navy
 210 channels, spectral resolution ~ 10 nm
 Wavelengths: 413 to 2,504 nm.
 spatial dimension - 320 pixels.
 Spatial resolution dependent on altitude (1 – 10 m)
DAIS (Digital Airborne Imaging Spectrometer)
European
spectral coverage between 450 and 2,500 nm
72 channels, three spectral intervals:
 400-1200 nm (bandwidth 15-30 nm);
 1,500-1,800 nm (bandwidth 45 nm);
 2,000-2,500 nm (bandwidth 20 nm).
 The gaps coincide with atmospheric absorption
bands

Satellite Sensors
Moderate Resolution Imaging Spectrometer
– Near-polar, 705 km sun-synchronous orbit

– Whiskbroom scanner
– 36 bands
• 20 bands from 400 to 3000 nm
• 16 bands from 300 to 15,000 nm
– Spatial resolution
• 250 m x 250 m (bands 1-2)
• 500 m x 500 m (bands 3-8)
• 1 km x 1 km (bands 8-36)
Hyperion
The Hyperion EO-1 sensor was launched in November 2000 by NASA with the
purpose of taking hyperspectral images from space in order to create
mineralogical mapping. Hyperion is a hyperspectral satellite sensor which
works in the spectral range 0.40…2.50 μm with 242 bands which have a
spectral resolution of about 10 nm and a spatial resolution of 30 meters, the
data is taken from an altitude of 705 km. Hyperion is a push-broom
instrument that takes pictures with a radiometric resolution of 8 bits, the
band having a w idth of 7.5 km and being perpendicular on the movement of
the satellite. The system used for taking images is formed of two
spectrometers: one working in the visible/near infrared (VNIR) (0.4…1 .0 μm)
and one in shortwave infrared (SWIR) (0 .9…2.5 μm). The data are calibrated
using both the radiation measured before the mission and when the images
are taken
• 220 spectral bands

• from 0.4 to 2.5 μm
• 30 meter spatial resolution.
• images a 7.5 km by 100 km imaged

CHRIS Compact High Resolution Imaging Spectrometer
 Part of the Proba satellite, launched 2001,

European Space Agency
 18 km x 18 km images
 18 meter resolution
 200 spectral bands
IMS-1 - HYSI
Indian Mini Satellite- 1 : Hyperspectral Imager
Launched April 2008

Chandrayan - HYSI
 64 contiguous bands in the VNIR, Spectral range - 0.4-0.95

μm region
 Spectral resolution of better than 15 nm
 Spatial resolution of 80 m,
 swath coverage of 20 km.

Future hyperspectral sensors
•Environmental Mapping and Analysis Programme (ENMAP) of DLR
Germany
Spectral range from 0.430 to 0.950 m (VNIR) and from 0.950 to
2.400 m (SWIR) with 184 channels, a swath width of 30 km at high
spatial resolution of 30 m and off-nadir (30◦) pointing feature for
fast target revisit (<3 days).
•PRISMA mission by Italian space agency
(700 km orbit, 20–30 m resolution, swath width of30–60 km, 0.4–
2.5 m continuous coverage with 10 nm bands) to be launched in
2012 by the Italian Space Agency
Hyper–multi spectral mission named HISUI (HyperspectralImager
SUIte), Japanese hyperspectral mission
57 VNIR bands (0.4–0.97 micometer at 10 nm resolution) and 128
SWIR bands (0.9–2.5 microm. at 12.5 nm resolution) at a 30 m
spatial resolution and 30 km swath
Preprocessing Hyperspectral Data

1. Atmosphere correction
2. Noise removal and dimensionality reduction

INTERACTION WITH THE ATMOSPHERE
Because hyperspectral data is collected some distance above

the target on the ground, the reflected solar illumination must
travel through the atmosphere.
Characteristics of the atmosphere can have a profound effect on

the incoming solar energy recorded by the hyperspectral
sensor.
The way light interacts with the atmosphere depends on many

factors: the types of particulates and gases, the amount of
atmospheric reflection, atmospheric absorption, and
atmospheric scattering. The atmosphere is a complex mixture of
particulates and gases. Particulates are small particles usually
less than 20 μm in diameter. Particulates greater than 20 μm are
not likely to stay in the atmosphere for long durations and tend
to settle to the ground rather quickly
Amount of Atmospheric Absorption
Atmospheric absorption is the removal of energy from solar

irradiance by conversion of the electromagnetic energy to
another form, usually thermal energy. Atmospheric absorption
occurs when a photon induces a molecular vibration, rotation,
or electron orbital transition to an alternate energy state.
The photon is absorbed by the constituent molecules of the

atmosphere, and only photons with specific energy levels can be
absorbed.

Molecular Absorption
Molecules can absorb electromagnetic radiation in three ways:
electronic transitions, vibration, and rotation. Electronic transition
requires the greatest amount of energy and involves the
promotion of electrons to higher energy levels. Vibration involves
the molecular bond between atoms and models it as a spring.
Rotation can be considered in the context of a simple diatomic
molecule of two atoms. The two atoms can rotate about their
center of mass.
Molecular absorption can be complicated by a combination of

mechanisms occurring simultaneously. For example, the energy
level of a molecule can be described by both rotation and
vibration.
Amount of Atmospheric Scattering
The surface area of the particulate and gases interacts with the light
and creates scattering or a redirection of the electromagnetic
radiation (EMR), which can cause a change in the distribution of the
EMR.
The three basic types of scattering are classified as Rayleigh, Mie, and non-
selective scattering. Rayleigh scattering occurs w hen the EMR interacts w ith the
minute particles or mole cules that are the components of the atmosphere,
primarily when the particles are much sm aller than the wavelength of the
incident flux. Mie scattering results when the wavelength of the incident EMR is
approximately equal to the size of atmospheric particles, such as aerosols, dust
particles, fossil fuel combustion products, and suspended sea salts. Nonsele ctive
scattering occurs when the suspended atmospheric particles are very large w ith
respect to the incident EMR; particles such as water droplets and ice crystals can
cause nonselective scattering

Target reflectance is a function of:
• Atmospheric irradiance (path radiance: R1)

• Reflectance outside target scattered into path
(R2)
• Diffuse atmospheric irradiance (scattered
onto target: R3)
• Multiple-scattered surface-atmosphere
interactions (R4)
Atmospheric Correction Models

Physics based Models
 Atmospheric REMoval (ATREM)
 Atmospheric Correction Now (ACORN)
 Fast Line-of-sight Atmospheric Analysis of Spectral
Hypercubes (FLAASH)
Semi-Empirical/Empirical Models

Atmospheric correction is the process of converting satellite
signals (at-sensor radiance) to ground reflectance
 Removes atmospheric and solar illumination effects

 Improves change detection
 Used with spectral library based classifiers
 Simplifies satellite data inter-comparisons
Different levels of atmospheric correction yield different

approximations of scene reflectance
 Planetary reflectance –no knowledge of atmosphere

 Ground reflectance using knowledge of atmosphere
 Ground reflectance using knowledge of atmosphere and
adjacency effects
Atmospheric Correction Measures
Compensation for the shape of the solar spectrum.

The measured radiances are divided by solar irradiances
above the atmosphere to obtain the apparent reflectance
of the surface.
Compensation for atmospheric gaseous transmission and
molecular and aerosol scattering enable apparent reflectances to
be converted to scaled surface reflectances
Scaled surface reflectance can be converted to real surface
reflectance after accounting for topographic effects or by
assuming that surfaces are lambertian

Flat Field Correction
Flat Field Correction (FFC) technique is used to normalize images to an
area of known uniform “flat” reflectance (Kruse et al., 2003b).
The method requires locating a large, spectrally flat, spectrally uniform

area in the data, usually defined as a Region of Interest (ROI). The
radiance spectrum from this area is assumed to be composed of
primarily atmospheric effects and the solar spectrum.
The average radiance spectrum from the ROI is used as the reference
spectrum, which is then divided into the spectrum at each pixel of the
image.
FFC normalizes the hyperspectral data to an area of known flat

reflectance, and derives relative reflectance from hyperspectral data.
The resultant apparent reflectance is comparable with laboratory
spectra.
Internal Average Relative Reflectance (IARR) calibration

technique is used to normalize images to a scene average
spectrum.
This is particularly effective for reducing imaging spectrometer

data to relative reflectance in an area where no ground
measurements exist and little is known about the scene (Kruse et
al., 2003b).
It works best for arid areas with no vegetation. The IARR

calibration is performed by calculating an average spectrum for
the entire scene and using this as the reference spectrum.
Apparent reflectance is calculated for each pixel of the image by
dividing the reference spectrum into the spectrum for each pixel.

Empirical Line Calibration
 Empirical Line Calibration technique is used to force image data
to match selected field reflectance spectra (Kruse et al., 1990).
 This method requires ground measurements and/or knowledge.
Two or more ground targets are identified and reflectance is
measured in the field.
 Usually the targets consi st of at least one bright and one dark
area. The same two targets are identified in images and average
spectra are extracted for Regions of Interest.
 A linear regression i s calculated between the field reflectance
spectra and the image radiance spectra to determine a linear
transform from radiance to reflectance for each band of the data
set.
 Gains and offsets calculated in the regression are applied to the
radiance spectra for each pixel to produce apparent reflectance
on a pixel-by-pixel basis.
 In this method, linearly transform the hyperspectral imaging
radiance values to ground reflectance
 This process i s equivalent to removing the solar radiance and the

atmospheric path radiance from the measured signal
R = G L +O (+  )
R calculated reflectance for band , O is the offset for band , L is
the at sensor radiance for band ,   is the error term from the
regression for band
 The relative methods are computationally fast, but the information

pertaining to the intervening atmosphere is not derived in these
approaches. These technique s are considered more of a
calibration tool than an atmospheric correction model.

Absolute Atmospheric Correction
Ideally, researchers would like to be able to calibrate their data to
absolute reflectance rather than relative reflectance without actually
having to make ground measurements.
Absolute atmospheric correction methods require a description of

the components in the atmospheric profile. The output of these
methods is an image that matches the reflectance of the ground
pixels with a maximum estimated error of 10 %, if atmospheric
profiling is adequate enough.
The advantage of these methods i s that ground reflectance can be

evaluated under any atmospheric condition, altitude and relative
geometry between sun and satellite.
The disadvantage is that the atmospheric profiling required for these

methods is rarely available.
Most of the atmospheric correction models like ATREM, ACORN,

FLAASH, etc. ba sically follow the radiative transfer model shown
below. though each model use s a slightly different version and the
FLAASH algorithm adds a term to account for adjacency effects.
Current atmospheric correction programs assume that the surface is horizontal

and has a Lambertian reflectance. This is be cause for real data we typically don’t
have enough information to make the topographic *cos(θ)+ correction. The end
result is called “scaled surface reflectance” or “apparent reflectance”. The scaled
surface reflectance can be converted to surface reflectance if the surface
topography is known.

MODTRAN radioactive transfer model
MODTRAN (MODerate spectral resolution atmospheric TRANSsmittance
algorithm and computer model) is developed by AFRL/VSBT in collaboration
with Spectral Sciences, Inc. The MODTRAN Code calculates atmospheric
transmittance and radiance for frequencies from 0 to 50,000 cm-1 (wavelength:
200nm to +∞) at moderate spectral resolution.
Except for its molecular band model parameterization, MODTRAN adopts all the
LOWTRAN 7 capabilities, including spherical refractive geometry, solar and lunar
source functions, and scattering (Rayleigh, Mie, single and multiple), and default
profiles (gases, aerosols, clouds, fogs, and rain).
The optical interaction effect between sun-atmosphere-surface and surface-

atmosphere- sensor is constituted three sets . It is showed that the radiance at
the sensor is composed of three different contributions: (1) radiation scattered
by the atmosphere into the viewing direction. (2) radiation reflected from the
target and directly transmitted in the viewing direction. (3) radiation reflected
from the background (surroundings) and diffusely transmitted to the sensor,
which is called the adjacency effect
Assumed that the surface is uniform and has a Lambertian reflectance, the radiance at a
downward looking aircraft sensor can be written in a simplified form as follow

ATREM – ATmospheric REMoval Program
ATREM is software developed by the University of Colorado for retrieving scaled
surface reflectance from hyperspectral data using a radiative transfer model
(Gao and Goetz, 1990; Gao et al., 1993; CSES, 1999).
First the solar zenith angle is derived based on the AVIRIS acquisition time,
date, and geographic location.
Atmospheric transmittance spectra are derived for e ach of seven atmospheric

gases [water vapor (H2O), carbon dioxide (CO2), ozone (O3), nitrous oxide
(N2O), carbon monoxide (CO), methane (CH4), and oxygen (O2)] using the
Malkmus narrow band model (Malkmus, 1967).
A water vapor “lookup table” is created by generating modeled spectra for

various water vapor concentrations, again using the Malkmus narrow band
model and estimating the 0.94 and/or 1.13 micrometer water vapor band
depths for each spectrum.

Band depths are determined using a ratio of the band center to the two band
shoulders. Water vapor is then estimated for e ach AVIRIS pixel by determining
the band depth and comparing to the modeled band depths in the lookup table.
The output of this procedure is an image showing the spatial distribution of
various water vapor concentrations for each pixel of the AVIRIS data.
Atmospheric scattering is modeled using the “6S” radiative transfer code, (Tanre
et al., 1986). Apparent reflectance spectra are obtained by dividing each AVIRIS
spectrum by the solar irradiance curve above the atmosphere (Kneizyx et al.,
1983) and using the water vapor image along with the other atmospheric
parameters in the radiative transfer model of Tiellet (1989).
The final results are a water vapor image and reflectance-corrected AVIRIS data
without use of ground spectral measurements. While the ATREM software
package is no longer supported and is not available to new users, many HSI data
users have and use ATREM. It demonstrates baseline atmospheric correction
capabilities. ATREM Version 3.1 was the last publicly released software and was
used for this research (CSES, 1999).

ATREM - EFFORT
Even though the ATREM correction is adequate for most analysis purposes,
the spectra usually still contain residual atmospheric and instrument effects,
which is difficult to direct comparison with spectral librarie s. To remove such
effect, empirical models can be used effectively like EFFORT (Empirical Flat
Field Optimized Reflectance Transformation).
The EFFORT correction takes advantage of the fact that a typical hyperpsectral
scene contains a number of featureless spectra because these spectra still
contain the noise effects, they can be used to characterize the nature of noise.
Feature less spectra are identified by calculating a Legendre Polynomial Fit to
each spe ctrum in the image and then determining which spectra are well
modelled by their corresponding polynomial.
Once the well modelled spectra are determined, then the gains and offset
required to explain the difference between modelled spectra and actual
spectra. A linear regression is used to calculate the gain and offset for each
band that will make ATREM featureless spectra look like the modeled spectra
These gains and offset are then applied to every spe ctrum in the
hypserspectral dat aset. The resulting spe ctrum contain all of the fine
absorption features.

ACORN: ATMOSPHERIC CORRECTION NOW
ACORN is a commercially-available, enhanced atmospheric model-
based software that use s licensed MODTRAN4 technology (Berk et al,
1999) to produce high quality surface reflectance without ground
measurements. The package provides an atmospheric correction of
Hyperspectral and Multispectral data measured in the 0.4 - 2.5
micrometer spectral range (AIG, 2001).
ACORN use s look-up-tables calculated with the MODTRAN4 radiative

transfer code to model atmospheric ga s absorption a s well as
molecular and aerosol scattering effects, converting the calibrated
sensor radiance measurements to apparent surface reflectance (AIG,
2001). The well mixed gases are constrained by the elevation and the
observation geometry.
Water vapor is e stimated from the data on a pixel-by-pixel basi s using

the water vapor absorption bands at 0.94 and/or 1.150 micrometers. A
lookup table for a range of water column vapor densities is generated
using MODTRAN4 and then fitted in a least-square s sense against the
imaging spectrometer data.
A key feature of ACORN is full spectral fitting to solve for the overlap of
absorptions between water vapor and liquid water in surface vegetation.
Visibility is estimated from the AVIRIS data using nonlinear le ast-squares
spectral fitting between the AVIRIS radiance spectra and MODTRAN modeled
radiance with the aerosol optical depth as the primary fitting parameter.
The two-way transmitted radiance and atmospheric reflectance are calculated

for each pixel using MODTRAN and the derived water vapor, pressure elevation,
and aerosol optical depth estimations.
Apparent surface reflectance is derived from the total upwelling spectral

radiance for a given atmosphere using a variant of the radiative transfer
equation
The principal outputs of ACORN are a water vapor image and a scaled surface
reflectance cube . An image showing an estimate of leaf-water is also optionally
produced. ACORN artifact suppression options include automated wavelength
correction, removal of noisy channels, and “polishing” of residual errors. The
latest version, ACORN 4.15 was used for this research.

FLAASH - Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes
FLAASH is a MODTRAN4-based atmospheric correction software package
developed by the Air Force Phillips Laboratory, Hanscom AFB and Spectral
Sciences, Inc (SSI) (Adler-Golden et al., 1999).
It provides accurate, physics-based derivation of apparent surface reflectance

through derivation of atmospheric properties such as surface albedo, surface
altitude, water vapor column, aerosol and cloud optical depths, surface and
atmospheric temperatures from HSI data.
FLAASH operates in the 0.4 – 2.5 micrometer spectral range. First, MODTRAN
simulations of spectral radiance are performed for various atmospheric, water
vapor, and viewing conditions (solar angles) over a range of surface reflectance s to
establish lookup tables for the atmospheric parameters of column water vapor,
aerosol type, and visibility for subsequent use.
Typically, the 1.13 micrometer water band is used to estimate water vapor, and a
ratio of in-band and out-of-band radiance values allows estimation of absorption
band depths for a range of water vapor column densities.
FLAASH also derives pressure altitudes by applying the same method to the
oxygen 0.762 micrometer absorption band. The radiance spectra are extracted
from the data and compared against the MODTRAN lookup tables on a pixel-
by-pixel basis to determine scaled surface reflectance.
FLAASH offers the additional option of correcting for light scattered from
adjacent pixels. Spatially averaged reflectance is used to account for the
“adjacency effect’’
FLAASH provides additional flexibility when compared to the other two

atmospheric correction program s in that it allows custom radiative transfer
calculations for a wider range of conditions including off-nadir viewing and all
MODTRAN standard aerosol models.

Column Water Vapor Amounts and Surface Temperatures for the
MODTRAN Model Atmospheres

Selecting an Aerosol Model
• Rural: Represents aerosols in areas not strongly affected by urban or
industrial sources. The particle sizes are a blend of two distributions,
one large and one small.
• Urban: A mixture of 80% rural aerosol with 20% soot-like aerosols,

appropriate for high-density urban/industrial areas.
• Maritime: Represents the boundary layer over oceans, or continents

under a prevailing wind from the ocean. It is composed of two
components, one from sea spray and another from rural continental
aerosol (that omits the largest particles).
• Tropospheri c: Applies to calm, clear (visibility greater than 40

km) conditions over land and consists of the small-particle component
of the rural model.
For more details on MODTRAN aerosol models, see Abreu and
Anderson (1996).

FLAASH includes a method for retrieving the aerosol amount
and estimating a scene average visibility using a dark pixel
reflectance ratio method based on work by Kaufman et al.
(1997).
The dark-land pixel-retrieval method requires the presence of

sensor channels around 660 nm and 2100 nm. A dark-land pixel
is defined to be one with a 2100 nm reflectance of 0.1 or less
and a 660:2100 reflectance ratio of approximately 0.45.
If the input image contains bands near 800 nm and 420 nm, an
additional check is performed, requiring the 800:420 radiance
ratio to be 1 or less, which eliminates pixels likely to be
shadows and water bodies.

e – Learning Material: Unit-3&4
Data Cubes and Spectral Vectors

A spectral cube is a three-dimensional array containing spatial
(image) information on the x and y axes and spectral information
on the z axis. Individual spectra, spectral maps, and full spectral
cubes can be created from a single spectral cube.
Spectral cubes display different stratified, graphical thicknesses in

a three-dimensional perspective that can be a useful tool for
overall image analysis.

As a result of spatial and spectral sampling, airborne hyperspectral
imaging (HSI) sensors produce a three dimensional (3D) data
structure (with spatial-spatial spectral components), referred to as a
data cube.
The 3D Hypercube Display emphasizes the high spectral content of
the hyperspectral image while allowing you to quickly examine any
of the individual wavelength bands. The 3D Hypercube Display
window portrays the hyperspectral image as a three-dimensional
“image cube”.
If we extract all pixels in the same spatial location and plot their
spectral values as a function of wavelength, the result is the average
spectrum of all the materials in the corresponding ground resolution
cell
The values of all pixels in the same spectral band, plotted in spatial
coordinates, result in a grayscale image depicting the spatial
distribution of the reflectance of the scene in the corresponding
spectral wavelength.

The top and right panels of the cube show the corresponding edge cells of each
wavelength band, with wavelength increasing toward the back of the cube.
Basic data-cube structure (center) in hyperspectral imaging, illustrating the

simultaneous spatial and spe ctral character of the data. The data cube can be
visualized as a set of spectra (left), each for a single pixel, or as a stack of
images (right), each for a single spectral channel
Spectral library
The spectral library is a collection of spectra of natural
and man made materials.
These libraries provide a source of reference spectra of

varieties of targets( minerals, rocks, vegetation species)
for remote identification for these targets.
The relative distance between sensor and target does

not affects the results of spectral library.

JHU(John Hopkins University) Laboratory
Minerals (0.4-15 micrometer)
Rocks (2-25 micrometer)
Manmade features (0.3-12.5 micrometer)

Lunar rock samples
USGS_Vegetation laboratory
Vegetation ( Visible-near infrared –shortwave domain)
Jet propulsion laboratory for rocks and minerals

Minerals( Visible-near infrared –shortwave domain)
USGS( United States Geological Survey) Spectral laboratory
Minerals( Visible-near infrared –shortwave domain)

Rocks( Thermal domain)

Importance of Spectral library development
Spectral signatures of minerals and few vegetation species are
unique. Spectral features help in identifying the elements /
targets.
Characterization of elements based on spectral signatures is

rapid and reliable method.
Spectral profiles of elements (say rocks and minerals) should

be preserved with every details (grain size and semi
quantitative mineralogy etc.).
Spectral features are as reliable as any other semi quantitative

chemical methods.
Reference for target detection-(Known to unknown)
Factors controlling the quality and information in spectra

External factors
 Properties of a spectrometer
 Wavelength within which measurements are being made
 BRDF(Influences the continuum of the spectral curve not

the spectral feature)
 Intensity of incident radiation( Lamp used in lab or sun

light)
Target/Intrinsic Factors
Atomic processes operative in the target or sample.
Texture(Grain size etc.)

External Factors
Properties of a spectrometer
Parameters that describe the

capability of a spectrometer:
(1) Spectral range
(2) Spectral bandwidth
(3) Spectral sampling,
(4) signal-to-noise ratio.
1) Spectral range: Spectral range is important to cover enough

diagnostic spectral absorption to solve a desired problem. a)
Ultraviolet (UV): 0.001 to 0.4 μm, b) visible: 0.4 to 0.7 μm, c) near-
infrared (NIR): 0.7 to 3.0 μm, d) the mid-infrared (MIR): 3.0 to 30
μm, and d) the far infrared (FIR): 30 μm to 1mm
2) Spectral bandwidth: Spectral band width is the width of an
individual spectral channel in the spectrometer. The narrower the
spectral Bandwidth, the narrower the absorption feature the
spectrometer will accurately measure.
3) Spectral sampling:
Spectral sampling is the distance in wavelength between the
spectral band pass profiles for each channel in the spectrometer
as a function of wavelength. The Nyquist theorem states that the
maximum information is obtained by sampling at one-half the
FWHM

Spectral resolution: narrowest spectral features that can be
resolved by a spectrometer(full width at half maximum FWHM)
4) signal-to-noise ratio (S/N). The S/N is dependent on the

detector sensitivity, the spectral band width, and intensity of
the light reflected or emitted from the surface being
measured.
Bidirectional reflectance-distribution function (BRDF)

Bidirectional reflectance-distribution function (BRDF), a mathematical function
“relating the irradiance incident from one given direction to its contribution to the
reflected radiance in another direction ”(Nicodemusetal.,1977). Importantly, BRDF
was defined as a conceptual property of the surface, at infinitesimally small angles

Wavelength domain for reflectance spectroscopic studies
a)Ultraviolet (UV): 0.001 to 0.4 μm,

b)visible: 0.4 to 0.7 μm,
c)near-infrared (NIR): 0.7 to 1.0 μm,
d)short-wave infrared(SWIR): 1.0 to 2.5 μm
e)the mid-infrared (MIR): 2.5 to 30 μm,
f)the far infrared (FIR): 30 μm to 1
NIR DOMAIN IS IMPORTANT FOR VEGETATION.
SWIR DOMAIN IS IMPORTANT FOR MINERALS

Intrinsic Factor(Target Factor)
Factors influence the spectral signature Changes in Composition
Reflectance spectra of muscovite showing band shifts

due to changing aluminum composition
Role of crystallinity
Subtle spectral
differences in the
kaolinite group
minerals near 2.2-μm.
KaoliniteCM9 is well
crystallized (WXL)
while KGa-2 is poorly
crystallized (PXL).
Spectral bandwidth is
1.9 nm and sampling is
0.95 nm.

Composition plus crystal structure
Comparison of calcite and dolomite continuum-removed

features. The dolomite absorption occurs at a shorter
wavelength than the calcite absorption.
Role of grain size
cba=Refle ctance spectra of pulverized sample(<100 micron) of

Carbonaceous kimberlite= Reflectance spectra of pulverized sample(60-
100 micron) of carbonaceous kimberlite. c= Refle ctance spectra of
consolidated Carbonaceous kimberlite.

Rock spectra vs Mineral spectra

Laboratory set up
where S is source, L is
fiber optic lens where H
refers height of the
measurement gunfrom
the sample top and d
indicates the field of
view (FOV)of the
measurement gun.

Calibration Panel
•Perfectly diffuse reflector.

•Performance of the calibration panel is important
•Maintenance of calibration panel
•Recalibrating the calibration panel.
Sample preparation
•Sample( Intact rock sample; size to be specified)
•Grain size of the powdered sample.
•Soil texture should be maintained.
•For few cases; specific grain sized sample is essential.
Pul verized Ki mberlite Rock sample

Ki mberlite Rock sample

Continuum Removal: Continuum is a imaginary line joins
highest albedo points in a spectral profile

Continuum-removed chlorophyll
absorptions for 8 vegetation types
showing that the continuum removed
features can show subtle spectral
differences.

Noise Estimation and dimensionality
reduction in Hyperspectral Data
Data Reduction
 Advances in data colle ction during past decade Resulting in information
overload
 Traditional statistical methods breakdown - Due to Increase in number of
Observations & Variables associated with each observation
 Dimension of the data is the number of variables – measured on e ach
observation
 High Dimensional data sets present many challenges
 Major difficulty with high dimensional data
 Not all measured variables are “important” for unraveling phenomena of
interest
Commonly used techniques are
Principal Component Analysis – PCA

Minimum Noise Fraction – MNF

Principal Components Transform
Different bands of spectral data are often highly correlated - contain similar
Information Based on statistical characteristics – data redundancy & correlation
between bands can be reduced
Principal Components Analysis - Is a widely used technique for dimensionality

reduction & data compression.
PCA Obje ctive is to reduce dimensionality (i.e. the number of bands) in the
data, and compress as much of the information in the original bands into fewer
bands “NEW" bands that result from this statistical procedure are called
components
Process attempts to maximize (statistically) the amount of information (or

variance) from the original data into least number of new components
Principal Components Analysis

Uses Eigen values determine the significance of PC
Data Reduction is in accordance with associated eigenvalues

First principal component shows the direction and length of the widest transect
of the ellipse, measures the highest variation within the data.
Second principal component is the widest transect of the ellipse that is

orthogonal (perpendicular) to the first principal component Second principal
component describes the largest amount of variance in the data that is not
already described by the first principal component In a two-dimensional
analysis, the second principal component corresponds to the minor axis of the
ellipse

In n dimensions, there are n principal components
Each succe ssi ve principal component: Accounts for a decreasing

amount of the variation in the data which is not already accounted
for by previous principal components
Although there are n output bands in a PCA. First few bands

account for a high proportion of the variance in the data in some
cases, almost 100%.
PCA i s useful for compressi ng data into fewer bands Use ful
information can be gathered from principal component bands

By seeing this plot we can see that eigen
values are high for initial few bands and
after eigen value number 6-7 (band 6-7)
eigen value is constant.
Eigen values have direct relation with the

variance of the data higher the eigen value
the variance is more and the information is
more.
Therefore for further processing only these

few bands will be sele cted. The first band
contains the largest percentage of data
variance and the second PC band contains
the second largets data variance and so on
the last bands appe ar noise as they
contain little variance

Issues of PCA
PCA maximizes total image variance with out making distinction

between Noise & Information
PCA images are ordered by data variance. If bands have

differing amounts of noise ?
Bands in a hyperspectral image have differing amounts of noise

Principal components may not show usual trend of steadily
increasing noise with increasing component number

Distinction between information & noise can be made
• Some sensor systems simultaneously acquire dark images.
• Dark Current images represent electronic noise present in the

data
• These dark images can be used as input for noise estimation
Absence of dark images -If we assume that
Signal at any point in the image is strongly correlated with the

signal at neighboring pixels while the noise shows only weak
spatial correlation
Near neighbor differences can be averaged to derive noise

value

Minimum Noise Fraction Transform
Shift Difference method
It is assumed that each pixel contains both signal and noise,
and that adjacent pixels contain the same signal but different
noise
“shift difference” is performed on the data by differencing

adjacent pixels and averaging the results to obtain the “noise”
value to assign to the pixel being processed
Best noise estimate is gathered using the shift-difference

statistics from a homogeneous area rather than from the
whole image
Minimum Noise Fraction transform (MNF) orders output components by

decreasing signal to noise ratio
MNF procedure first estimates the noise in each im age band using the
spatial variations in brightness values
Then applies two successive principal component transforms
First uses the noise estimates to transform the dataset to a coordinate

system in which the noise is uncorrelated and is equal in each component
Then a standard principal components transform is applied to the noise-

adjusted data, with output components ordered by decreasing variance
Procedure produces a component set in which noise levels increase

uniformly with increasing component number
Low-order components should contain most of the im age information and

little image noise

MNF IS PERFORMED SEGREGATING DATA INTO VNIR &
SWIR SEGEMENTS

MNF is useful in determining inherent dimensionality of image
data by segregating noise and reduce computational
requirements for subsequent processing

Pixel Purity Index
The set of low-order MNF components provides a “distilled”
version of the hyperspectral image that can be used to rapidly
identify relatively “pure” image spectra for use as spectral
endmembers in the further processes.
The Pixel Purity Index (PPI) operation is the first step in identifying
these endmember spectra.
Pixel Purity Index

The "Pixel-Purity-Index" (PPI) is a means of finding the most
"spectrally pure," or extreme, pixels in multispectral and
hyperspectral images. See Boardman et al . (1995).
PPI calculates a spectral purity score for each n-dimensional pixel

in the original data by generating random unit vectors (called
skewers), so that all pixel vectors are projected on to the skewers
and the ones falling at the extremes of each skewer are counted.
After many repeated projections to different skewers, those
pixels that count above a certain cut-off threshold are declared
‘‘pure’’
A Pixel Purity Index (PPI) image is created in which the DN of

each pixel corresponds to the number of times that pixel was
recorded as extreme

A two-dimensional illustration
of how the Pixel Purity Index
identifies potential extreme
image spectra (large red dots).
Endmembers represent pure pixels supposedly representing one feature type

only and these can be collected from image itself

Using PPI images in a n-dimensional visualiser for
endmembers collection
Spectra can be thought of as points in an n-dimensional

scatterplots, where n is the number of bands.
The coordinates of the points in n-space consist of “n” values that

are simply the spectral radiance or reflectance values in each band
for a given pixel.
The distribution of these points in n-space can be used to estimate

the number of spectral endmembers and their pure spectral
signatures.
n-Dimensional Visualizer
To assess the results of the PPI operation we will use the n-
Dimensional Visualizer tool.
This tool allows you to create and view an n-dimensional scatterplot

of spectra for an area in the hyperspectral image or MNF component
set. You can view the scatterplot from different viewpoints and
rotate the plot manually or automatically in real time.
You designate the selected area by drawing a polygon in the

Hyperspectral Image window. You can use this tool to investigate
the spectral properties of different materials in the image and to
search for image cells with extreme spectral values that might
represent pure endmembers

Using PPI images in a n-dimensional visualiser for
endmembers collection
The n-Dimensional visualizer provides an opportunity for interactive

selection of the endmembers in n-space. The n-D visualizer is used
in conjunction with the Minimum Noise Fraction Transform (MNF)
and Pixel Purity index (PPI) tools to locate, identify and cluster the
purest pixels and most extreme spectral responses in a data set.
The n-Dimensional visulizer allows for interactive rotation of data in

n-D space, selection of groups of pixels into different classes
(Boardman, 1993; Boardman and Kruse, 1994). Here in this study
this procedure was followed to isolate different group of pixels
representing different endmembers. The selected classes were
exported to Region of Interest (ROI) and used as input for further
spectral processing.

The n-D Space — Where Many Algorithms Operate
Each HSI spectrum (or pixel) is an n-D vector that
can be represented as a single point in n-D space.
n-D space is actually where many of our algorithms
operate.
Spectrum (or pixel )  1 , 2 , 3 , 4 , 5 , 6 , 7 ,..., n 

T
1.0
0.9
Ref lectiv ity, 
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.4 0.8 1.2 1.6 2.0 2.4
Wav elength (mm)
Four (A-D) Equivalent Notations/Representations

A B
1.00
...imagine an n-D
0.90 hy perspace...
, Band b
0.80
Reflectance, 
0.70
0.60 Spectrum s 1
0.50
0.40
0.30
0.20
0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50
Wavelength (micrometers)
, Band a
C (0.11, 0.23, 0.30, 0.25, 0.16, 0.27, 0.31, 0.37,...,)
D
0.11î  0.23 ĵ  0.30k̂  0.25l̂  0.16m̂  0.27n̂  0.31ô  0.37p̂  ... 

Some HSI Scatter Plots; Spectra as Points in ‘Hyperspace’
Classification Techniques
1. Full Pixel
2.Sub Pixel
Full Pixel or per pixel classification
Full-pixel analysis methods attempt to determine whther one or more target

materials are abundant within each pixel in the image on the basis of the
spectral similarity between the training (reference) pixel and target
(unknown) spectra Spectral Angle Mapper
1. Spectral Feature Fitting
2. Support Vector Machines

Spectral Angle Mapper or SAM
The spectral angle is the angle between any two vectors
originating from a common origin. The magnitude of the angle
indicates the similarity or dissimilarity of the materials—a smaller
angle correlates to a more similar spectral signature.
This method is relatively insensitive to changes in illumination on

the target material because changes in light will impact the
magnitude but not the direction of the vector. A poorly
illuminated target will cause the points to be plotted closer to the
origin
SAM: n-D Geometry

Angular Distance Metric (Spectral Angle Mapper or SAM)
Assume a two band spectral remote sensing system. Each two point
‘spectrum’ is a point in Band b vs. Band a space.
A 2D scatterplot with 2 spectra:
The angle, , between the two

lines connecting each spectrum
Spectrum s 2
Band b
(point) to the origin is the angular

Spectrum s 1 separation of the two spectra.
Smaller angular separations in-
 dicate more similar spectra.
Band a

Variation in illumination
affected more strongly
when the Maximum
Likelihood and Minimum
Distance were used. The
results of the
classification
deteriorated fast while
the Spectral Angle
Mapper and the Spectral
Correlation Mapper
were better in that case.

Spectral Correlation Mapper
X = Image Spectrum (DN value) of the pixel in all spectral bands

_
X = Mean of the DN values of the same pixel over all spectral
bands
Y= Reference Spectrum of a class in all spectral bands

_
Y = Mean of the reference spectra of a class over all spectral
bands
1. Aim is to classify pixel X
2. Given DN value of pixel X for every band
3. Use
to determine R for each class
4. Assign the pixel to the class where the value of R is

maximum

Support Vector Machine
SVM performs classification by constructing an N-dimensional hyper plane
that optimally separates the data into two categories
Support vector Machines are good candidates for remote sensing

classification in particular hyperspectral image classification in many ways
1. SVM can work well small training data set

2. SVM perform with high classification accuracy for data having hundreds
of dimension as in hyperspectral images
3. The structure of SVM is less complex even with high dimensional data
The SVM aim to maximize the margin between two classes of interest by
placing linear separating hyperplane
The SVM uses structural risk minimization (SRM). The Empirical risk
minimizes the misclassification error and whereas structural risk
minimization minimizes the probability of misclassification.

Support Vector Machine
A set of features that describes one case (i.e., a row of predictor value s) is
called a vector. So the goal of SVM modeling is to find the optimal
hyperplane that separates clusters of vector in such a w ay that cases w ith
one category of the target variable are on one side of the plane and cases
with the other category are on the other size of the plane.
The vectors near the hyperplane are the support vectors. The figure below
presents an overview of the SVM process.
The points lying on the two parallel hyperplanes are called Support Vector
(SV). The m argin is defined as the Euclidian distance between the Support
vector and Optimal separating hyperplane (OSH). Maximization of margin
ensures better generalization capability for the test data
SVM are based on the structural risk minimization (SRM) which aims at
minimizing the upper bound of the expected error over the whole data set.
The advantage of SVM is its ability to work with smaller training data sets,
SVM are becoming preferred classification method to classify hyperspectral
imagery
A Two-Dimensional Example
Before considering N-dimensional hyperplanes, let ’s look at a simple 2-

dimensional example. Assume we wish to perform a classification, and our data
has a categorical target variable with two categories.
Also assume that there are two predictor variables with continuous values. If we
plot the data points using the value of one predictor on the X axis and the other
on the Y axis we might end up with an im age such as shown below. One
category of the target variable is represented by rectangles while the other
category is represented by ovals.

In this ide alized example , the cases with one category are in the lower left
corner and the cases with the other category are in the upper right corner; the
case s are completely separated. The SVM analysis attempts to find a 1-
dimensional hyperplane (i.e. a line) that separates the cases based on their
target categories. There are an infinite number of possible lines; two candidate
lines are shown above. The question is w hich line is better, and how do we
define the optimal line.
The dashed lines drawn parallel to the separating line m ark the distance
between the dividing line and the closest vectors to the line. The distance
between the dashed lines is called the margin. The vectors (points) that
constrain the width of the margin are the support vectors. The following figure
illustrates this.
An SVM analysis finds the line (or, in general, hyperplane) that is oriented so
that the margin between the support ve ctors is maximized. In the figure above,
the line in the right panel is superior to the line in the left panel.
If all analyses consisted of two-category target variables with two predictor

variables, and the cluster of points could be divided by a straight line, it would
be easy. Unfortunately, this is not generally the case, so SVM must de al with (a)
more than two predictor variables, (b) separating the points with non-line ar
curves, (c) handling the cases where clusters cannot be completely separated,
and (d) handling classifications with more than two categories.
Flying High on Hyperplanes

In the previous example, we had only two predictor variables, and we were
able to plot the points on a 2-dimensional plane. If we add a third predictor
variable, then we can use its value for a third dimension and plot the points in
a 3-dimensional cube. Points on a 2-dimensional plane can be separated by a
1-dimensional line. Similarly, points in a 3-dimensional cube can be separated
by a 2-dimensional plane.

As we a dd additional predictor va riables (attributes), the data points can be represented
i n N-dimensional space, a nd a (N-1)-dimensional hyperplane can separate them.
When Straight Lines Go Crooked

The s implest way to divi de two groups is with a straight line, flat plane or an N-
di mensional hyperplane. But what i f the points are s eparated by a nonlinear region such
a s s hown below?
In thi s case we need a nonlinear divi ding line.
Ra ther than fitting nonlinear curves to the

da ta , SVM handles this by using a kernel
functi on to map the data into a different space
where a hyperplane ca n be used to do the
s eparation.
The kernel function may transform the data i nto a higher dimensional space to
ma ke i t possible to perform the s eparation.

The concept of a kernel mapping function is very powerful. It allows SVM models to
perform separations even with very complex boundaries such as shown below
Many kernel mapping functions can be used – probably an infinite number. But a few
kernel functions have been found to work well in for a wide variety of applications.
The default and recommended kernel function is the Radial Basis Function (RBF).

Spectral Feature Fitting
To match target and reference pixel spectra by examining spe cific absorption
features in the spectra (continuum removed spectrum)
User spe cifies a range of wavelengths within a unique absorption feature exists
for the chosen target
Reference (training) spe ctra are then compared to the target spectrum using
two measurements:
 The depth of the feature in the target is compared

to the depth of the feature in the reference, and
 The shape of the feature in the target is compared

to the shape of the feature in the reference (using a
least-squares technique)

Continuum Removal
 Continuum Removal normalizes reflectance spectra to allow comparison of
individual absorption features from a common baseline
 Continuum is a convex hull fit over the top of a spectrum utilizing straight -
line segments that connect local spectra maxima
 Continuum is removed by dividing it into the actual spectrum for each pixel
in the image
 The resulting image spectra are equal to 1.0 where the continuum and the
spectra match and less than 1.0 where absorption features occur
Kaolinite reflectance spectra Continuum removed spectra
Band depth
The apparent depth of an absorption feature, relative to the surrounding
continuum in a reflectance or emittance spectrum
Rb is the reflectance
at the absorption-
band center (the
minimum in the
continuum-removed
feature), and
Rc is the reflectance
value of the
continuum at the
wavelength of the
band center
If the features one is looking for have weak absorption Continuum removal
suppresses them

Spectral Unmixing - Sub-Pixel analysis
Mixed pixels are generated if the size of the pixel includes more
than one type of terrain cover
Measured spectral radiance of a pixel is the integration of the
radiance reflected from all the type of terrain covers within the
ground instantaneous field of view (GIFOV)
Spectral mixing is inherent in any finite-resolution digital
imagery
Mixed pixel problem not well addressed with multispectral data
because distinctions with limited band numbers were clear to
differentiate classes
Unmixing has particular relevance for something like mineral
mapping where abundance of minerals is desired parameter

a: Presence of small or sub-pixel
targets
b: Presence of boundaries of
discrete parcels
c: Gradual transition between
land cover classes
d: Contribution of areas outside
the area represented by a pixel

Spectral unmixing Is a procedure by which the measured spectrum of
a mixed pixel is decomposed into a collection of constituent spectra,
or endmembers, and
a set of corresponding fractions, or abundances, indicating the

proportion of each endmember present in the pixel.

Linear -Constrained Unmixing
If we can claim to know the spectral reflectance or radiance for
the materials potentially in each pixel (i.e., the endmember) we
can write M simultaneous linear equations in N unknowns (i.e.,the
fractions are the only unknowns).

Two very different types of unmixing are typically used: Using
“known” endmembers and using “derived” endmembers.
Using known endmembers, one seeks to derive the apparent fractional

abundance of each endmember material in each pixel, given a set of “ known” or
assumed spectral ndmembers.
These known endmembers can be drawn from the data (averages of regions
picked using previous knowledge), drawn from a library of pure materials by
interactively browsing through the imaging spectrometer data to determine what
pure materials exist in the image

Matched filtering
The technique is a unique approach to spectral mixture modelling

in that it does not require knowledge of the spectral signatures of
other component materials (Boardman 1998). A type of unmixing
in which only user chosen targets are mapped
Unlike Complete Unmixing, to find the spectra of all endmembers
in the scene to get an accurate analysis (hence, this type of analysis
is often called a “partial unmixing” because the unmixing
equations are only partially solved)
Matched Filtering “filters” the input image for good matches to the
chosen target spectrum by maximizing the response of the target
spectrum within the data and suppressing the response of
everything else (which is treated as a composite unknown
background to the target)
Pixel value in the output image is proportional to the fraction of
the pixel that contains the target material
Matched filtering
MF vector equates to target abundance estimations that range

from 0 to 100% (Mundt et al. 2007). Spectra that closely match the
training spectrum will have a score near one while background
noise will have a score near zero. Any pixel with a value of 0 or less
would be interpreted as background (i.e., none of the target is
present).
One potential problem with Matched Filtering is that it is possible
to end up with false positive results
One solution to this problem that is available in ENVI is to calculate
an additional measure called “infeasibility”. Which is the method
called MTMF
Is a hybrid method based on the combination of the matched filter
method (no requirement to know all the endmembers) and linear
mixture theory

This technique may find some “false positives” that shows high values in
terms of matching score for rare materials which is taken care of by an
output “infeasibility” image to the results.
The infeasibility image is used to reduce the number of “false positive s”

that are sometimes found using m atched filtering technique. Pixels w ith
a high infe asibility are likely to be matched filter false positives. Correctly
mapped pixels will have a high matched filter score and a low infeasibility
value where as Pixels with a high matched filter result and high
infeasibility are “false positive” pixels and do not match the target. The
infeasibility values are in noise sigm a units which vary in digital number
scale with matched filter score and indicate the feasibility of the matched
filter results (Harsanyi & Chang, 1994; Chen & Reed, 1987).
Mixture Tuned Matched Filtering technique combines the best part of

the linear spectral unmixing model and the statistical matched filter
model while avoiding the drawbacks of e ach model. From matched
filtering it inherits the advantage of its ability to map a single known
target without knowing the other background endmember signatures,
unlike traditional spectral mixing models.

MTMF (Mixture-Tuned Matched Filtering )
 Is a hybrid method based on the combination of the
matched filter method (no requirement to know all
the endmembers) and linear mixture theory.
 The results are two images:
 a MF score image with 0 to 1 (1 is perfect match), and
 A infeasibility image, the smaller the better match.
 Infeasibility is based on both noise and image statistics and
indicates the degree to which the Matched Filtering result is a
feasible mixture of the target and the background. Pixels with
high infeasibilities are likely to be false positives regardless of
their matched filter value.
 Use 2-D scatter plot to locate those pixels in an image.

Spectral Derivative Analysis
Benefited from the high spectral resolution of hyperspectral data, the reflectance could
construct a comparatively continuous curve, and thus the derivative method is likely to
be used for data analysis.
From one hand, the derivative processing could enhance the small variations of spectral
curve (Li, 2006). And from the other hand, the derivatives could also be relatively less
sensitive to the spectral variations of sunlight and skylight (Tsai and Philpot, 1998), and
also eliminate background signals (Becker et al, 2005).
The First Derivative Reflectance (FDR) and Second Derivative Reflectance (SDR) were
calculated according to Eq. (1) and Eq. (2). For FDR method,

HYPERSPECTRAL REMOTE SENSING
Hyperspectral remote sensing

in mineral exploration
J.Saravanavel - Hyperspectral-4

A mineral is an element or chemical compound that is normally
crystalline and that has been formed as a result of geological
processes.
a naturally occurring solid chemical substance formed through

bio-geochemical processes, having characteristic chemical
composition, highly ordered atomic structure, and specific
physical properties.
The main factors that influence the reflectance characteristics of

a substance are:
 Chemical composition
 Texture / particle size
In geology, we refer to rocks as aggregates of minerals. Hence,

reflectance depends on mineral composition and degree of
weathering
Wavelength Range
APPLICATION OF HYPERSPECTRAL REMOTE SENSING

Requirement of Spectral resolution is determined by the width of the
occurring absorptions. For Soils and Geological investigations mostly in the
near and short wave infrared a 10mm resolution is necessary
Soils and Rocks

Charge Transfer Absorptions
Occur mostly in the visible region of the spectrum, and are caused by light at
certain wavelengths, causing ele ctrons to be transferred between atoms. • Eg.
Fe3+ and Fe2+. Light at the proper wavelength causes an electron to be
transferred from a Fe2+ atom to a Fe3+ atom
Reflectance spectra of Iron

oxides and hydroxides
display absorption
features (T shaped
symbols) due to
crystal-field effects and Fe-
O charge transfer.
Iron ore (Hematite) shows

maximum reflectance at 0.7μm and
strong absorption at 0.85μm.
Reflectance spectra of two olivines showing the change in band position and
shape with composition. The 1-µm absorption band is due to a crystal field
absorption of Fe 2+.
The Fo 29 sample has an FeO content of 53.65%, while the Fo 91 sample has an
FeO content of 7.93%.
The mean grain size is 30 and 25 µm respectively. The 1-µm band position varies
from about 1.08 µm at Fo 10 to 1.05 µm
J.Saravanavel at Fo 90 (King and Ridley, 1987).
- Hyperspectral-4

two pyroxenes showing
the change in Fe2+ -
absorption band
position and shape with
composition (from Clark
et al., 1993b).
Diopside is CaMgSi2O6, but some Fe2+ substitutes for Mg.
Bronziteis (Mg,Fe)SiO3 with mostly Mg.
The 1-µm versus the 2-µm band position of a pyroxene describes the
pyroxene composition J.Saravanavel - Hyperspectral-4
ELECTRON TRANSITION ABSORPTIONS
In atoms with an incomplete electron shell, light at the proper

wavelength can bump electrons in to different positions in the
shell.
These absorptions tend to be narrower than the charge transfer

absorptions, & the wavelengths of the absorptions are controlled
by the type of atom and the position & variety of its neighbours.
This feature is very useful in geology, where the arrangement of

atoms in a mineral is well defined.
Since subtle variations in the position of the band centre are very
important, it is necessary to have many narrowly spaced bands
to take full advantage of this feature.

Reflectance spectra of the i ron oxi de hema ti te (Fe2O3) and iron hydroxide goethite
(FeOOH, from Cla rk et al., 1993b). The intense charge-tra nsfer band in the UV (< 0.4 µm) is
"satura ted" in reflectance, so onl y fi rst surfa ce (specula r) reflection is seen in these spectra.
The 0.9-µm and 0.86-µm abs orption fea tures a re due to Laporte-forbidden transitions (e.g.
Morris et al , 1985; Sherman, 1990 a nd references therein). The absorption a t 2.7-3 µm is
due to tra ce water in the samples ., and in the case of goethite, the OH. The goethite
s pectrum is offset upward 0.2 units.J.Saravanavel - Hyperspectral-4
Reflectance spectra of Sulfur, S, (top) and cinnabar, HgS, (bottom) showing

conduction bands in the visible (from Clark et al., 1993b).

Minerals containing chemically bound water have distinctive absorption
features (T shaped symbols) for the H-O-H bond stretching transitions
Reflectance spectra of clay minerals and micas contains distinctive

absorption features (T shaped symbols) due to the bending of Al-OH and
Mg-OH bonds. Such features help in discriminating between minerals of
hydrothermal & sedimentaryJ.Saravanavel
process.- Hyperspectral-4

Reflectance spectra of carbonates contains distinctive absorption features (T
shaped symbols) due to vibrational transitions related to C-O bonds. Most
distinctive features is near 2.35 microns, w hich helps to discrim inate between
carbonates and clays.
Silicates have different structures.

Hence, the positions of Si-O bond
stretching trough and peaks occur at
different positions for different
silicate minerals

Spectra of non-
silicates in mid IR
and TIR show
different patterns
of absorption from
silicates. Thus, it
is possible to
discriminate
lithologies by
sensing
emitted thermal
radiation

OH Examples. The sharper OH-related absorption bands
allow smaller band shifts to be measured. These bands can
be so sensitive that it is possible to distinguish between the
isochemical end-members of the Mg-rich serpentine group,
chrysotile, antigorite, and lizardite (King and Clark, 1989a,
and Figure).
Mustard (1992) calibrated changes in the 1.4-µm and 2.3-
µm absorptions in the tremolite-actinolite solid solution
series; example spectra of the 1.4-µm absorptions are
shown in Figure.
High spectral resolution reflectance spectra of the first
overtone of OH in talc, tremolite, actinolite, crysotile,
lizardite, and antigorite. The three sharp absorption bands
in talc, tremolite and actinolite are caused by Mg and Fe
ions associated with the hydroxyls, causing small band
shifts. The Fe:Fe+Mg ratio can be estimated. In chrysotile,
lizardite and antigorite, the absorptions change with small
structural differences even though the composition is
constant
The strong and sharp OH features have proven
particularly diagnostic of clay mineralogy

Al in Muscovite
More recently, subtle shifts have been found in muscovite series with aluminum
composition. As elements substitute for aluminum in the crystal structure, the
crystal becomes slightly distorted relative to no substitutions.
This cause s slight changes in Al-O-H bond lengths and thus shifts absorption
band position. In this case , the shift of the 2.2-µm absorption appears
continuous with composition.
Reflectance spectra of muscovite showing band shifts due to changing aluminum

composition J.Saravanavel - Hyperspectral-4
Identifying
Minerals We
can use all of
these
absorption
features to
determine the
chemical
composition
from a spectral
reflectance
curve

Role of Remote sensing in mineral exploration
• Mapping regional lineaments and structural trends
• Mapping local fracture patterns
• Detecting hydrothermally altered rocks
• Providing basic geologic data
GEOLOGICAL GUIDES FOR PROSPECTING
There are several geological guides for prospecting that can be observed
on remotely sensed data. These include the following:
1) Stratigraphical-Lithological
2) Geomorphological
3) Structural
4) Rock Alteration
5) Geobotanical
Hyperspectral > Another Empirical Data

Layer for Prospectivity Analysis
Hyperspectral
Mineral Map

Hydrothermal Wall rock alteration
The wall-rock alteration is quite common in case of hydrothermal
deposits.
The nature of mineralizing solution like its chemical character,

temperature and pressure as well as character and kind of wall
rock decide about the nature and intensity of alteration.
Sericitization, kaolinisation and silicification are the common forms
of alteration in the wall rock.
High temperature minerals like tourmaline, topaz and amphibole

may develop. Basic and ultrabasic igneous rocks are serpentinized
accompanied by the production of epidote and chlorite.
Types of Hydrothermal Wall rock alteration

Advanced argillic - characterized by the clays dickite, kaolinite and pyrophyllite
(all hydrated alum inum silicates) and quartz. Sericite may be present as well as
alunite and tourmaline . Alteration involves the extreme leaching of cations,
especially the difficult to leach alkalis and calcium , and the concentration of H+.
This type of alteration is characteristic of many epithermal precious metal
deposits and a smaller number of mesothermal deposits such as Butte, Montana.
Potassic alteration - char acterized by se condary kspar + biotite. Anhydrite may be

present, but its susceptibility to solution generally results in its dissolution in
near surface environments. Because it is characterized by common silicates,
potassic alteration is often difficult to detect. Pyrite and minor chalcopyrite and
molybdenite are the only ore minerals associated with this alteration.
Sericitization (Phyllic) - characterized by the assemblage quartz + sericite + pyrite.

Generally the most common form of alteration. Sulfide s present, in addition to
pyrite, include chalcopyrite, bornite and a variety of less common copper
sulfides. Generally the result of oxidation of Fe.

Argillic - characterized by kaolinite + montmorillonite. Somewhat similar to advanced
argillic alteration, but with a lesser degree of leaching of cations. Also unlike advanced
argillic alteration which is associated with vein type deposits, argillic alteration is more
closely associated with disseminated deposits, porphyry coppers in particular. Sulfides are
less common in association with this alteration type.
Propylitic - characterized by the assemblage chlorite + epidote + calcite. Albite as well as

other carbonates may be present. Due to presence of the green minerals chlorite and
epidote this zone is usually easily recognizable by its color. Associated sulfides include
pyrite, copper sulfides, galena, sphalerite and a host of complex arsenides. Often this zone
can be quite large and is useful during mineral exploration. Unlike the previous types
above which are characterized by leaching of cations this zone seems to represent the
addition of cations.
Silicification - characterized by quartz or chert. Can be added by solutions as is the case in

many low temperature deposits or the result of complete leaching of all cations plus
aluminum.
Dolomitization - addition of magnesium to limestone to form dolomite. Common in

Mississippi Valley type deposits.
Other alteration types:
Feldspathization - kspar + albite, forms in the deep zones of some porphyry

copper deposits.
Greisenization - tourmaline + topaz + cassiterite + various tungsten-bearing

minerals. Common form of alteration on association with porphyry tin deposits.
Fenitization - characterized by nepheline, alkali feldspar and Na-bearing

amphiboles. Hematization - characterized by secondary hematite.
Bleaching - not characterized by any specific mineral assemblage, but rather a

color change between altered and unaltered rock. Generally the result of
oxidation of Fe.

Two examples of spectral characteristics of common mineral
groups is as follows.


Alteration Minerals Identifiable by Hyperspectral data

 Clays: Kaolinite, Dickite, Halloysite, Pyrophyllite, Illite-Smectite,
Montmorillionite
 Micas: White mica (Sericite), Illite, Phengite, Muscovite,
Paragonite, Biotite, Phlogopite, Pyrophyllite, etc.
 Chlorites: Variations in Fe, Mg content
 Other Phyllosilicates: Serpentines (Lizardite, Antigorite), Fuchsite,
Talc, Nontronite
 Amphiboles: Tremolite, Hornblende, Actinolite, Anthophyllite
 Carbonates: Cal+cite, Dolomite, Ankerite, Siderite
 Sulphates: Alunite, Jarosite, Gypsum
 Tourmaline: Fe-tourmaline, Tourmaline
Major Absorptions:
• Al-OH: clays, micas
• Fe-OH: chlorites, serpentines
• Mg-OH: serpentines, chlorites, micas,
epidote, some amphiboles

Possible Mineral Deposits by Hyperspectral Alteration Mapping
Porphyry Deposit Model – Hydrothermal

Alteration Zones > Mineral Mapping
Figure: Illustrated deposit

m odel of a porphyry copper
deposit (modified from
Low ell and Guilbert, 1970).
A. Mineral Alteration Zones
B. Mineralization/Ore Shells

Epithermal Deposit Model –
Hydrothermal Alteration Zones
Diagram from IT AM Gold by the Minerals Council of Australia
VMS Deposits – Hydrothermal Alteration

Figure: Representativ e examples
of alteration zoning in ancient
v olcanogenic massiv e sulfide
deposits.
A, Hydrothermal alteration in the
Bathurst district after Goodfellow
(2007).
B, Alteration in the mafic, ophiolitic
Turner-Albright, OR, deposit
(Zierenberg and others, 1988)
show s an alteration pattern related
to replacement mineralization in a
porous and permeable hyaloclastite
pile w ith general similarities to
alteration at TAG and the
Galapagos Rift.
C, Chisel Lake deposit in the Snow
Lake district, Manitoba, w here
alteration has undergone
amphibolite-grade regional
metamorphism (Galley and others,
1993).
D, Semi-conformable alteration
zones (Hannington and others,
2003) that are clearly discordant to
regional metamorphic isograds in
the Blake River Group, Noranda
volcanic complex, w estern Abitibi
subprovince, Ontario. [Fe, iron; Mg,
magnesium; Fm., formation]
From Shanks et al., 2010 (USGS Publication: Volcanogenic Massive Sulfide Occurrence Model)

VMS Deposits – Hydrothermal Alteration
 Studies of alteration assemblages in these continental hydrothermal settings led to a
series of commonly recognized alteration zones: potassic, argillic, phyllic, and
propylitic, with distinct mineralogy and, respectively. Decreasing intensity of alteration
developed away from the vein or pluton
 Bonnet and Corriveau (2007) retained some of these classification terms (Table 11–1)
and used some of the assemblage names for VMS deposits, but substituted sericitic for
phyllic and, like other researchers, added chloritic as an important alteration zone in
subseafloor settings. Advanced argillic is a special type of alteration that forms in
highly acidic, high sulfidation-state conditions characteristic of near-seafloor (or near-
surface) oxidation of SO2 or H2 S to produce sulfuric acid.
From Shanks et al., 2010 (USGS Publication: Volcanogenic Massive Sulfide Occurrence Model) RPM. Workshop – CRSS, Ottawa, June, 2012
Kimberlite Pipe – Diamond Exploration

Simplified version of a Kimberlite pipe from
Haw thorne (1975)
 Diamonds are brought to the surface from

the mantle in a rare type of magma called
kimberlite and erupted at a rare type of
volcanic vent called a diatreme or pipe.
 Kimberlite is a gas-rich, potassic ultramafic
igneous rock that contains the minerals
olivine, phlogopite, diopside, serpentine,
calcite, and minor amounts of apatite,
magnetite, chromite, garnet, diamond, and
other upper mantle minerals.
 Some of these minerals can be
Hy Vista, 2011
differentiated using hyperspectral imagery
 VNIR-SWIR: olivine, phlogopite, serpentine,
calcite
 TIR: calcite, diopside, garnet, apatite, other
silicates
 Challenge: pipes are small in size/diameter;
mafic rocks with lower spectral signal
alteration minerals
RPM. Workshop – CRSS, Ottawa, June, 2012

Reflectance and Composition
 SWIR Spectral Characteristics
(Hermann et al., 2001) RPM. Workshop – CRSS, Ottawa, June, 2012
Alteration Composition Trends

& Mineralization – VHMS (Cu + Au)
 PIMA of pul v. dh samples, 15-50 m s pacing
 Ba ckground/distal = s l. phengitic (Fe-Mg)
mi ca >(2200-2210 nm) - (qtz+chl+ser±carb)
 Proxi mal alteration = Pa ragonitic, sodic
mi ca >(2194-2198 nm) – (py+qtz+ser±pyrop)
 Sys tematic decrease in Al-OH wavelength
towa rds the mi neralized zone
> Vector to Ore; Pyrophllite ID also sig.
(Herrman et al., 2001)


SINED Hope Bay Mapping Results – Block A
Deposit: Doris North
Exploration
Known
Targets
Geology f rom Sherlock et al., (2004)
• High Al Illite coincident with v esicular basalts adjacent to mineralized zones Legend:
> sericite identif ied with intense alteration halos proximal to mineralization SINED mineral
Potential results
• chlorite associated with gabbros, peripheral massiv e basalts & cultural f eatures
> chloritie has been observ ed in rinds and f ine v arioles in basalts ! NUMIN (GSC)
?
Au occurrences
SINED Survey: L. Wickert J.Saravanavel - Hyperspectral-4
SINED Hope Bay Mapping Results – Block A
 Hope Bay – Discovery Area

 Block A; NW of Doris
 One reported Au occurrence
 Reported carbonate alteration, but
!
? not mapped
 Significant abundance of mapped
High Al illites
 Chlorites & pyrophyllite also noted
 Geology not well mapped in this
area? (e.g. Sherlock, 2003 does not
cover area)
!
? NUMIN (GSC) Regi onal Geology,
Au occurrences
from Sherlock et al., (2004)
SINED Survey: GSC, L. Wickert J.Saravanavel - Hyperspectral-4


Schematic cross
section of USGS
mineral deposit. This
interpretive model
for the Creede
caldera system,
Yellowstone National
Park region exhibits
the alteration pattern
typical for acid-
sulfate systems (from
Mosier and others,
1986).
Figure shows the spectra for clay minerals, calcite, alunite, and siliceous sinter. The
unique absorption features of interest generally are in the 2.2- to 2.4-μm-wavelength
region. S iliceous sinter, as used here, is a spectral group of S iO 2 minerals including
opal, chalcedony, and cristobalite derived from hydrothermal fluids and (or) the
decomposition of rhyolite. The identification of these minerals uses a hydration-
absorption feature that is present in some silica at the 2.2- to 2.4-μm-wavelength region

Figure shows the spectra for iron oxide (hematite), iron hydroxide (goethite), and iron
sulfate (jarosite). Shifts in the absorption-feature minima of these minerals in the 0.5- to
0.6-μm region are responsible for the colors of minerals as the eye sees them—yellow
(jarosite), brown (goethite), and red (hematite). Other features used for identification are
in the 0.42- and 0.9-μm wavelengths J.Saravanavel - Hyperspectral-4

Mammoth Hot
Springs is a large
travertine deposit
that was precipitated
from water that
percolated through
underlying limestone
bedrock. Mammoth
Hot Springs and
Terrace Mountain, an
inactive travertine
deposit to the
southwest, have
strong carbonate
spectral features that
map as calcite.









Hyperspectral Remote
Sensing of Vegetation
Hyperspectral Remote Sensing of Vegetation

More specifically hyperspe ctral Remote Sensing, originally used for detecting and
mapping minerals, is increasingly needed for to characterize, model, classify, and
map agricultural crops and natural vegetation, specifically in study of
(a)Species composition (e.g., chromolenea odorata vs.

imperata cylindrica);
(b)Vegetation or crop type (e.g., soybeans vs. corn);
(c)Biophysical properties (e.g., LAI, biomass, yield, density);
(d)Biochemical properties (e.g, Anthrocyanins, Carotenoids,
Chlorophyll);
(e)Disease and stress (e.g., insect infestation, drought),
(f)Nutrients (e.g., Nitrogen),
(g)Moisture (e.g., leaf moisture),
(h)Light use efficiency,
(i)Net primary productivity

Spectral Wavelengths and their Importance in the Study of
Vegetation Biochemical properties
The reflectance
spectra with
characteristic
absorption
features
associated with
plant
biochemical
constituents for
live and dry grass
leaves from a senesced
birch (Betula),
ornamental beech
(Fagus) and healthy and
fully senesced maple
(AcerLf, Acerlit)
illustrating Carotenoid
(Car), The reflectance
spectra with
characteristic absorption
features associated with
plant biochemical
constitutents for live and
dry grass (Ad t d f Hill
[13]) Anthocyanin
(Anth), Chlorophyll (Chl),
Water and Lignocellulose
absorptions

LEAF AREA INDEX
Leaf area index (LAI) is defined as the total area of one-sided green leaves in
relationship to the ground below them (FAO 2005; Jensen 2007). Because LAI
directly quantifies the vegetation canopy structure, it is highly related to
diverse canopy processes, including water interception, photosynthesis,
evapotranspiration, and respiration. Thus, LAI information can be used in
various terrestrial monitoring and modeling applications to help quantify the
processes.
CHLOROPHYLL CONCENTRATION
The chlorophyll concentration in vegetation is m ainly responsible for the
absorption of photosynthetically active radiation (PAR). A high degree of
inter-correlation between canopy photosynthetic rates, LAI, absorbed PAR,
and chlorophyll concentrations is normal in natural ecosystems (Boegh et al.
2002; Jensen et al. 1998, 2002). The two optimum spectral regions for sensing
the chlorophyll absorption characteristics of a leaf are believed to be in the
blue (450 to 520 nm) and red (630 to 690 nm) portions of the spectrum. The
former region is characterized by strong absorption by carotenoids and
chlorophylls, whereas the latter is characterized by strong chlorophyll
absorption ( Jensen 2005; SimsJ.Saravanavel
and Gamon - Hyperspectral-4
2002).
BIOMASS
The amount of dead and live biomass is directly related to the productivity of
vegetation (e.g. in agricultural crops). In situ biomass measurement generally
involves measuring the biom ass for various parts of the plant (e .g. root, stem,
leaf ).
WATER CONTENT
Because water present in the spongy mesophyll of a plant absorbs much of the
energy in the mid-infrared spectral region, remote sensing techniques have been used
to identify leaf water content. Hence, as the water content of vegetation increases,
the reflectance generally decreases in the mid-infrared regions. For example, Figure 5
illustrates how the spectral reflectance of Bahiagrass changes according to various
irrigation treatments (Garcia-Quijano 2006).

SOIL NUTRIENT AVAILABILITY
The productivity and quality of vegetation are affected by soil nutrient

availability, which in turn influences leaf nutrient concentration (Townsend et
al. 2003). Leaf nutrients include nitrogen, phosphorus, potassium, calcium , and
magnesium . Nitrogen has been explored by many scientists using hyperspectral
remote sensing techniques because it is a major nutrient of vegetation. Other
nutrients, however, have had relatively less exploration because leaf nutrient
levels vary by differences in the physical structure of canopies and there may be
no strong linear relationship between spectral signature-derived values and the
nutrient levels. Phosphorus and potassium, critical to vegetation growth, are
used in various mechanisms, such as fat formation, energy transfer, cell
division, and seed sprouting ( Jokela et al. 1997; Milton et al. 1991).
Spectral reflectance curves of Bahiagrass subjected to various irrigation treatments

measured using an in situ ASD FieldSpec3 JR spectroradiometer (Garcia-Quijano 2006).



Hyperspectral Data on Tropical Forests
Factors Influencing Spectral Variation over Tropical Forests
1. Biochemistry (e.g., plant pigments, water, and structural carbohydrates): Leaf

reflectance in the visible spe ctrum is dominated by absorption features created
by plant pigments, such as:
chlorophyll a (chl-a): absorps in 410-430 nm and 600-690 nm;

chlorophyll b (chl-b): absorps in 450-470 nm;
carotenoids (e.g., β-carotene and lutein): peak absorption in
wavebands <500 nm; and
anthocyanins.
Lignin, cellulose, protein, Nitrogen: relatively low reflectance and
strong absorption in SWIR bands by water that masks other
absorption features
However, dry leave s do not have strong water absorption and reveal overlapping
absorptions by carbon compounds, such as lignin and cellulose , and other plant
biochemicals, including proteinJ.Saravanavel
nitrogen,starch, and sugars
- Hyperspectral-4
2. Structure or biophysical (e.g., leaf thickness and air spaces): of

leaves, and the scaling of these spectral properties due to
volumetric scattering of photons in the canopy;
3. Nonphotosynthetic tissues (e.g., bark, flowers, and seeds);
and
4. Other photosynthetic canopy organisms (e.g., vines,
epiphytes, and epiphylls) can mix in the photon signal and vary
depending on a complex interplay of species, structure,
phenology, and site differences






Hyperspectral Vegetation Indices (HVIs)


HVIs: Biophysical, Biochemical, Pigment, Water, Lignin and cellulose, and Physiology


Hyperspectral Remote Sensing of Vegetation
1. Hyperspectral narrowbands when compared with broadbands data

can significantly improve in:
1.1. Discriminating\Separating vegetation and crop types and their species;
1.2. Explaining greater variability in modeling vegetation and crop
biophysical, yield, and biochemical characteristics;
1.3. Increasing accuracies (reducing errors and uncertainties) in
vegetation\land cover classification; and
1.4. Enabling the study of specific biophysical and biochemical properties
from specific targeted portion of the spectrum.
2. About 33 narrowbands, in 400-2500 nm, provide optimal information

in vegetation studies. These waveband centers are identified in this
study. A nominal 3 to 5 nm wide bandwidth is recommended for all
wavebands;
3. Advances in methods and approaches of hyperspectral data

analysis in vegetation studies.



Hyperspectral Remote Sensing in SOIL
Soil is a complex system with highly variable in physical and chemical

properties (Ben-Dor et al., 1999). Both chemical and physical properties
have spectral responses and affect the shape and nature of a soil
spectrum.
Raw data, first-, and second-derivatives each provide valuable

information that can be analyzed separately or combined using
multivariate statistical methods or data mining techniques. Soil
constituents have unique absorption features in these wavelength
regions due to overtones related to stretching and bending vibrations in
molecular bonds such as C–C, C–H, N–H and O–H (Dalal and Henry,
1986).
This comprises the physical basis for the application of remote sensing
to map soil properties. Soil chemical chromophores (specific absorption
of spectrally active group) are materials in a soil system that absorb
incident radiation in discreet energy levels.
Soil has three major chemical chromophores: mineral (mostly clay and
iron oxides), organic matter and water (Ben-Dor, 2002).

The spectral features of clay minerals are associated with
oventones and combination modes of fundamental vibrations of
functional groups in the infrared region.
In general, three major spectral regions are active for clay: 1.3 -
1.4, 1.8 - 1.9, and 2.2 - 2.5 μm (Ben-Dor, 2002).
This is mainly based on specific absorption of spectrally active

groups such as Fe, OH- in water and minerals, CO3, Al, (Mg ) OH,
SO4 in minerals, and any others in organic matter.
Iron oxides can change soil color significantly due to selective

light absorption in the visible range. Soil iron content is highly
correlated with the absorbance in the spectral region 0.6 to 1.0
μm.
Soil organic matter (SOM) consists of plant and animal residues in

different stages of decomposition. SOM has strong influence on soil
reflectance, absorbing light through the VIS-NIR-SWIR (visible-near
infrared-shortwave infrared) region.
Ben-Dor et al. (1997) observed decreased soil albedo across the VIS-
NIR-SWIR region as the organic matter aged.
They suggested that the reflectance slope between 450 and 638 nm
may be useful for identifying both the degradation condition and the
parent material status of the organic matter and the reflectance slope
between 680 and 800 nm may be used as a general parameter for
monitoring organic matter maturity.
Moreover, organic matter was found to be correlated with the spectral

region 0.5 to 1.2 μm (Mathews et al., 1973) and that the region 0.9 to
1.22 μm is good for mapping soil organic matter (Beck et al., 1976).

Three major forms of water exist in soil: hydration water incorporated
into the lattice of the minerals, hygroscopic water adsorbed on soil
surface areas as a thin layer and free water filling soil pores (Ben-Dor,
2002).
Like SOM, moisture decreases soil reflectance in the VIS-NIR-SWIR

region. Spectra of a mixed system of water and minerals, exhibit OH
absorption features at around 0.95 μm (very weak), 1.2 μm (weak), 1.4
μm (strong) and 1.9 μm (very strong).
soil properties can be estimated from spectra of moisture free

soil more accurately than those of moist soil.
Spectra of oven-dried soil with

different content of organic matter
(A); Moisture effect on reflectance
spectra of soils with similar
amounts of organic matter and
different moisture content (B)

Spectral response pattern of soil is generally governed by a number
of factors. The properties of soils that govern their spectral
reflectance are colour, texture, structure, mineralogy, organic
matter, free carbonates, salinity, moisture and the oxides/
hydroxides of iron and manganese.
Chemical compositions of the soil influences spectral signature of

soils through the absorption processes. In near infrared (NIR) and
middle infrared (MIR) domain, absorption feature of soil
components in solid phase originate primarily from the vibrations of
bounded nuclei.
In addition to vibrations, molecular rotation and transition may

occur in the pores where gas and water molecules reside, which also
results in higher absorption in MIR region. Soil water exhibits
absorption peaks at about 1450 nm, 1880 nm and 2660 nm (Hoffer
1978).

Electronic transition processes require much higher energy levels

which causes absorption of light at lower wavelengths. Transition
elements (Fe, Mn, Ti etc.) explain most of the decrease in
reflectance observed towards the lower wavelength spectral
domain (Hunt & Salisbury 1976; Hunt 1980).
Organic matter absorbs strongly in shorter wavelength as well as

in the infrared region due to the presence of various functional
groups and conjugate bonds (Hoffer 1978; Hunt 1980).

Soil texture refers to relative proportion of sand, silt & clay and
affects the spectral reflectance of the soils due to its influence on
water holding capacity and the size of soil particles.

Spectral sensitivity analysis of Soil organic Carbon
Spectral features of soil

Structural soil crusts are relatively thin, dense, somewhat continuous layers of non-
aggregated soil particles on the surface of tilled and exposed soils. Structural crusts
develop when a sealed-over soil surface dries out after rainfall or irrigation.
This soil structural crust affect the some soil properties, such as infiltration, runoff and
soil erosion
A fi eld overview of
Ha mra soil after
exposure to a
na tural rainstorm.
Note the significant
col or differences
between the crusted
(l i ght) and non
crus ted (dark)
pl owed s oils.

Based on measurements of the three
soils and their corresponding spectral
fractions (crust, sole, and bulk, it is
postulated that an albedo
sequence occurs, moving from one
fraction to another.
All crusts are characterized by relatively

high reflectance values across the entire
spectral region, whereas the sole
fractions show relatively lower values.

Spectral signatures with the absorbance spectra of inorganic phosphates: the
peaks at 1.4, 1.9 and 2.2 μm and the absorptions at 2.1 and 2.3 μm.
HYPERSPECTRAL Remote sensing of water quality

Hyperspectral remote sensing can be used to estimate water
quality parameters such as suspended sediments, turbidity, and
chlorophyll
The use of remote sensing to quantify suspended sediment has

been studied extensively
Suspended sediments increase the energy reflected from

surface waters in the visible and nearinfrared spectrum. The
specific amount of increase depends on sediment type, texture,
color, water depth, and viewing conditions (Ritchie et al. 2003).
Although reflectance increased at all wavelengths, Ritchie et al.
(1976) concluded that wavelengths between 700 and 800 nm
were most useful for quantifying suspended sediment
concentration

Ritchie et al. (1974) developed an empirical approach to estimate suspended sediments.

The general forms of these empirical equations are
where Y is the remote sensing measurement (i.e., radiance , reflectance,

energy) and X is the water quality parameter of interest (i.e., suspended
sediment, chlorophyll). A and B are empirically derived factors. In empirical
approaches statistical relationships are determined between measured
spectral/thermal properties and measured water quality parameters.

Related to suspended sediment concentration is turbidity, a
measure of the degree to which light transmitted through the
water is scattered by suspended particles. Although turbidity is
also influenced by the quantity and type of organic particulates
(e.g., algae) in the water, suspended sediments are usually the
dominant particulate matter in surface water, and therefore are
the main cause of turbidity. The relationship between sediment
concentration and turbidity is influenced by the particle size
distribution of the sediment
Han (1996) achieved a correlation of 0.95 between turbidity and

the difference in reflectance at 710 and 720 nm for samples
obtained from a reservoir.
Several approaches used to estimate chlorophyll with remote

sensing were reviewed by Ritchie et al. (2003). Multispectral
approaches have not been successful in waters with high
suspended sediment concentrations because these broad-
wavelength data cannot successfully discriminate chlorophyll
where the spectral signal is dominated by sediment.
However, research has shown that narrow-band reflectance data

at the “red edge” of the visible spectrum can estimate
chlorophyll in the presence of high suspended sediment
concentrations.
This was particularly apparent when data from multiple

measurement dates were combined (derivative r = 0.82; ratio r =
0.55). Shafique et al. (2003) used the 705/675 nm reflectance
ratio to quantify chlorophyll a and reported similar results (r =
0.71 and 0.72) for two Ohio rivers.

Algae/Chlorophyll
Moni toring the concentrations of
chl orophyll (algal/phytoplankton)
i s necessary for managing
eutrophication i n
l a kes (Ca rlson 1977). Remote
s ensing has been used to measure
chl orophyll concentrations
s pa tially a nd temporally.
As wi th suspended sediment
mea surements, most remote
s ensing s tudies of chlorophyll i n
wa ter a re based on empirical
rel ationships between
ra di ance/reﬂectance in narrow
ba nds or ba nd ratios a nd
chl orophyll


E-Learning Material: Hyperspectral Remote Sensing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

E-Learning Material: Hyperspectral Remote Sensing

Uploaded by

Copyright:

Available Formats

6 Year Integrated M. Tech.

PAPER CODE: MTIGT0901

HYPERSPECTRAL REMOTE SENSING

e-learning Material – Hyperspectral Remote Sensing

1. Introduction: Understanding hypersectra, Multispectral vs hyperspectral remote

1. American Society of Photogrammetry, Manual of Remote Sensing (2nd Edition), ASP

e-learning Material – Hyperspectral Remote Sensing

e – Learning Material: Unit-1

 Since the beginning of time, light was thought to be pure and

 For longtime, thought that the light to be a bullet-like particle

e-learning Material – Hyperspectral Remote Sensing

 These discoveries of the interrelations among color, frequency,

Differs from conventional remote sensing in that it

e-learning Material – Hyperspectral Remote Sensing

The simultaneous acquisition of images of the same area in

Hyperspectral sensing (imaging spectrometry) takes remote

Ultraspectral ("beyond hyperspectral") is used to describe

Although no such sensor has been designed with this

The energy E of any photon is related to its frequency as follows,

e-learning Material – Hyperspectral Remote Sensing

Electromagnetic spectral regions (Sabines, 1987)

e-learning Material – Hyperspectral Remote Sensing

Atmospheric Transmittance: Windows for Remote Sensing

The drop toward the ultraviolet is due to scattering and strong

However, these spectral regions can be used in the laboratory

e-learning Material – Hyperspectral Remote Sensing

Radiation heat transfer is the exchange of thermal radiation energy between

l = εreflected + εabsorbed + εtransmitted

e-learning Material – Hyperspectral Remote Sensing

Because most solid bodies are opaque to thermal radiation,

The ratio of the actual emissive radiation E to the emissive

e-learning Material – Hyperspectral Remote Sensing

Reflectance is the ratio of reflected energy to incident energy.

Varies with wavelength

In climatology and remote sensing, reflectivity is commonly

Reflectivity R can be expressed as:

Reflected vs. Emitted Energy

e-learning Material – Hyperspectral Remote Sensing

Angle of Incidence = Angle of Reflectance

e-learning Material – Hyperspectral Remote Sensing

The portion of the electromagnetic spectrum sought for

Principles of Spectral Radiometry

A radiometer is a device used to measure the radiant flux or

The most important characteristics of a radiometer are spectral

e-learning Material – Hyperspectral Remote Sensing

These measures are used to characterize both emission from

Radiance characterizes total emission or reflection, while spectral

e-learning Material – Hyperspectral Remote Sensing

The term “irradiance” is used when the electromagnetic radiation

All of these quantities characterize the total amount of radiation

When this is done for a radiation incident on a surface, it is called

Radiant flux or radiant power is the measure of the total power

e-learning Material – Hyperspectral Remote Sensing

NASA AVIRIS Cuprite, NV, HSI Data, (1995)

e-learning Material – Hyperspectral Remote Sensing

e-learning Material – Hyperspectral Remote Sensing

Resampled to Landsat TM7 Bands

e-learning Material – Hyperspectral Remote Sensing

Low Spectral Resolution

e-learning Material – Hyperspectral Remote Sensing

Limitations and Issues of Hyperspectral Data

Very sensitive to noise

Difficult to interpret the spectral signatures of an “impure” pixel