Principles of Spectroscopy - Ground Truth; The "Multi" Concept; Imaging Spectroscopy - Remote Sensing Tutorial -
Principles of Spectroscopy

Much of this and the next two pages (until AVIRIS is discussed), has been condensed and reworded from a thorough review by Dr. Roger N. Clark of the U.S. Geological Survey, entitled Spectroscopy of Rocks and Minerals, and Principles of Spectroscopy, which you can access in its entirety at a site on the Internet. We took several of the illustrations in this summary from that source. Also, it may help to refresh your memory of the theory of spectroscopy covered on page I-2a in the Introduction. We start this page by discussing several basic concepts associated with the type of spectrometers used to analyze light for its constituent wavelengths; spectrometers used in chemical analysis are capable of separating and identifying specific wavelengths that are discrete and often unique to substances (such as elements) that give off EM radiation when excited.

The vital component of any spectrometer is its spectral dispersing device. These instruments have a physical means of spreading radiation composed of differing photon energies (thus a range of frequencies or their inverse, wavelengths) laterally onto a stretched linear display.

The simplest device is an optical glass prism whose cross-section is a triangle. A white (polychromatic) light beam arrives at one side of the prism.

Dispersion of white light by a prism.

At this glass surface, the rays with different wavelengths bend according to their response to the refractive power of the glass. The degree of bending varies with wavelength, which means that the index of refraction also differs for the range of wavelengths. For example, in crown glass the index is 1.528 for violet and 1.514 for red. If the beam strikes one side of a right-triangle prism at an angle, the light rays slow and bend (refract) to differing extents. When they emerge at the opposite side, they bend again. The net effect is to spread the (visible) light rays according to their effective indices of refraction into a continuous geometric color pattern (see pages I-2 and I-4). Light consisting of just a few discrete wavelengths again will bend differentially to emerge at specific angles characteristic of each wavelength.

A diffraction grating disperses light according to a different mechanism. The grating can be metal or glass, on which are ruled (cut) fine grooves of straight, parallel lines that are extremely close-spaced (e.g., 15,000 per inch), so that the spacing between pairs is roughly equivalent to the wavelengths in the Visible-NearIR region. Each line, analogous to a slit, causes polychromatic light to diffract (bend) at angles that depend on wavelength. For the array of closely-spaced lines (each pair separated by a distance, d), we apply the Diffraction equation: n λ = d sinΘ , where n refers to simple integers (1,2,3...) that establish orders (overtones), λ is wavelength, and Θ (Theta) is the diffraction or bending angle for that set of conditions. The diagram below shows a close view of a few of the lines in the grating. The formula in this diagram differs from the more general one above but is related ("a" is the line spacing; "k" refers to the order of spread)

The geometry of a diffraction grating.

Thus, a light ray striking the grating undergoes spreading at various angles, according to the wavelengths contained within it. Some wavelengths have larger numbers of photons (more intense) than others, so that we can recast a plot of reflected light energies as reflectance versus wavelength, hence yielding the type of spectral curves we have examined in the previous two pages.

An emission spectrometer, used to analyze material compositions, takes light of discrete wavelengths, representing excitation states of different chemical elements, through a slit and then a prism or diffraction grating onto a recording medium, such as a photographic plate. The wavelength-dependent bending reproduces a series of lines (repeated images of the slit) at varied spacings, whose wavelengths, we can measure, and thus identify the particular elements in the sample. The sample is usually heated in a flame or electric arc to force electrons into higher energy states, because light of given wavelengths is emitted according to the quantized energy when the electrons transition to lower states. Remember Planck's equation E = hf, where f is the frequency, discussed in the tutorial Introduction.

Spectroscopy as it applies to remote sensing is the science of measuring the spectral distribution of photon energies (as wavelengths or frequencies) associated with radiation that may be transmitted, reflected, emitted, or absorbed upon passing from one medium (vacuum or air) to another (material objects). In much of the Tutorial so far, our concern has been with reflectance spectroscopy. The exception - Section 9 in which objects emit radiation in response to the temperatures they assume when heated (usually after absorbing radiation from the Sun). Heating will always cause an object or material to experience some sort of radiative emission. High temperatures, such as brought on by solar fusion or by high voltage electrical excitation, produce diagnostic spectra that identify chemical elements and other substances when the emitted radiation is dispersed through prisms or gratings. Typical emission spectra for different elements can be displayed in spectrograms such as those shown here (in absorption spectra within the Visible region the lines appear black against a colored background):

Some of the emission spectral lines that identify by their position in the EM spectrum the elements specified.

Imaging spectroscopy is the special case in which spectral characteristics and variations of one variable tie to two additional variables (the spatial dimensions, given by x and y positions), to generate color composite images (pictures), ratio and principal-components images, and classification maps. In particular, images that represent the effects of diagnostic absorption bands can be produced to show specific spatial variability of certain material features that one or more such bands discretely identify.

When illumination (either polychromatic, like sunlight, or monochromatic, such as a laser beam) strikes a material, the electromagnetic radiation will likely partition into one or more components that behave differently. Some of the radiation directly reflects. If the material is transparent, the bulk of the radiation passes through but undergoes a change in direction according to the differences in indices of refraction between the material(s) and the external medium (usually air; or water). If the material is translucent or, more commonly, opaque, fractions (wavelength-variable) of the radiation may penetrate. When it penetrates, its rays undergo refraction, but some rays ultimately reflect. Of the fraction absorbed, some converts to heat, so that the temperature of the object rises, causing an increase in emissions, detectable as thermal radiation.

If the material is granular or polycrystalline, light that reflects will strike a number of surfaces (associated with grain or crystal boundaries), meeting individual surfaces at different angles of incidence and thus scattering the radiation at various angles. The light may bounce around, or back and forth, from several such surfaces before finally leaving as scattered beams. If the object's surface is smooth (mirror-like), a significant fraction reflects at an angle related to the angle of incidence. But, most surfaces tend towards some degree of roughness, so that the percentage or proportion of light reflected directly to the observer (an eye or an instrument) notably decreases. The fraction of radiation that is absorbed also controls the degree of scattering.

For a medium such as air, which also contains CO2 and other gases, water (usually as vapor or tiny droplets), and particulates, some scattering occurs. Hence, we have blue skies from a high degree of scattering at blue wavelengths, and red sunsets from additional scattering at longer wavelengths, leaving the red radiation less affected, i.e., transmitted as red.

But, air transmits or absorbs the bulk of the radiation, as indicated by this figure, with most of the minima (absorption bands) due to the presence of carbon dioxide, water, and oxygen.

Relative transmission of sunlight (with wavelength-dependent absorption) of the oxygen, carbon dioxide, and water vapor components of the atmosphere.

In the Visible-NIR range (VNIR), water ice and dry ice (solid CO2) give characteristic spectral curves, as shown here:

Comparison of spectral curves for water ice and carbon dioxide (dry) ice in the Vis-NIR spectral range.


Over most of this range, the dry ice remains highly reflective but displays a prominent set of absorption bands around 2 µm. Water ice reflects at shorter wavelengths, but its reflectance diminishes beyond about 1 µm. Note that there are several broad absorption bands that reduce reflectances to very low values. Liquid water tends to absorb well over most of the range, reflecting slightly more in the greens and blues.

While moderate to high reflectances are needed to produce the light tones in pictorial images, the wavelength-dependent absorption bands are the features in a spectral plot that commonly aid in identifying the materials that have bands (narrow to broad) that center at specific wavelengths. In the visible region, the reflected wavelengths control the observed color(s). Thus, a bright green color of an opaque material implies strong reflectance at green wavelengths and near total absorptance in the reds and blues. If, instead, there is also notable red reflectance, yellows to oranges would result. An absorbing medium affects the intensity of incoming radiation according to Beer's Law:

I = I0e-kx,

where I0 is the intensity of the incident radiation, e is the natural log base (2.71828...), k is a constant that depends on absorption as a function of the complex index of refraction (which takes into account the role of the extinction coefficient, K), and x is the depth of penetration. The bandwidth and depth of any given absorption depends on many factors, one of which is the spectral composition of the illumination.

To illustrate how we can use absorption bands plus reflectance levels to distinguish chemically similar materials, we now look at the spectra for two minerals: Hematite (Fe2O3) and Goethite (FeOOH). The first spectral curves, obtained with a spectrometer in a laboratory environment, cover the spectral intervals (ranges) between 0.3 and 1.0 µm (VNIR, for visible and near-infrared) and 1.0 to 2.5 µm (SWIR, for Short-Wave InfraRed).

Spectral curve diagram for Hematite and Goethite obtained from a spectrometer in a laboratory environment.

In this plot, the Goethite curve has been raised about 0.2 reflectance units above the Hematite curve to allow comparison. We use this offset procedure to separate multiple spectra that have similar reflectance levels. In this case, the two curves nearly coincide, if we place Goethite at its actual values. The absorption band at 0.86 µm for Hematite and 0.90 µm for Goethite permit us to distinguish the two spectrally but only with a high resolution spectrometer. These two minerals would not differentiate, if a broad-band system, such as the Multispectral Scanner Band 4 on Landsat, were the observing sensor. The OH molecule in the Goethite has a slight influence on its curve at about 2.4 µm. When we examine these two minerals in the mid-IR (thermal) range, a notably different response clearly separates them.

 Differences and similarities of the spectral curves for Hematite (iron oxide) and Goethite (a close relative, with the OH ion in its composition) over a spectral range that includes parts of the thermal IR used in remote sensing.

Here, there is no offset, so the overall reflectance of Hematite is greater. Note the dual narrow absorption bands for Hematite around 3 µm. A paired absorption band for Goethite near 6 µm is distinct from the single band for Hematite at 7 µm. At longer wavelengths, including two in the 8-14 µm interval available to thermal sensors, Hematite shows several shallow absorption bands (sometimes called "troughs," as contrasted to "peaks"). As we shall see later with several more examples, these MIR spectra can contain varied and definitive absorption features of considerable utility in distinguishing between and within classes of materials. Unfortunately, no one has developed yet a fine resolution instrument like AVIRIS (operating between 0.4 and 2.5 µm) for air/space platform use.

One more example emphasizes the power of detailed spectra in discriminating similar materials. We present offset curves covering a portion of the SWIR range for several minerals in the Kaolinite family of clays.

Spectral curve diagram for several minerals in the Kaolinite family of clays from a portion of the SWIR range.

WXL refers to well-crystalized, while PXL refers to poorly-crystallized. Although the gross expression of the absorption features is very similar among the four samples, slight shifts in the absorption band near 2.2 µm and other fine curve shifts suggest that under exceptional circumstances we could distinguish the members of this clay group, but only with difficulty by AVIRIS.

Absorption bands play a key role in defining the spectral curves for organic matter, such as vegetation. Consider this general curve (lighter line width) depicting the VNIR spectrum for a healthy oak leaf.

Variations in spectral response (reflectance) for an Oak leaf examined by a laboratory spectrometer, with differences due to water content.

At longer wavelengths, its pigments and cellular matter absorb light. Water bands also have a notable effect. Chlorophyll absorption dominates in the visible region, removing red and blue reflectances, leaving green as the dominant spectral wave range. The sharp rise in reflectance at 0.7 µm, continuing well beyond 1.1 µm, is largely the result of the walls of multiple cells reflecting the light. The second curve, rendered in heavier line weight, describes the spectrum of an oak leaf that is now dried and brown.

The next illustration shows four spectral curves, each for a particular vegetation type, and each, of the upper three, offset by 0.05 units from the one below. These were sampled from the hyperspectral image of Summitville, Colorado, displayed on page 13-10. In general, these plots are nearly identical, with variations mainly in the depths of individual absorption bands.

Four spectral curves of vegetative material of similar organic composition; these have been plotted with slight offsets (0.05 mm) to show by separation the similarity of the curve structure but there are small differences in reflectances owing to different plant shapes, etc.

Although difficult for the eye to detect and distinguish, there are real differences in equivalent absorption bands that allow us to make separations. The absorption band at 0.7 µm is a case in point. We need to use special processing to single out small differences.

A procedure that facilitates making this distinction is continuum-removal. The continuum consists of the so-called, "background absorption," which is in essence an extrapolation of the baseline of the general curve (fits a smoothed curve to the general trend so as to extend across the base of absorption bands). This local reduction specifies the continuum and is determined by mathematic manipulation of absorption coefficients by a subtraction process.

The depth of an absorption band, D, is:

D = 1 - Rb/Rc

where Rb is the reflectance at the bottom (trough center point) of a band and Rc is the continuum base.

The result for the above four vegetation plots (and several other crop types) is a set of continuum-removal curves that show slight to moderate differences in relative reflectances at a minimum centered on 0.68 µm. At least four of these crops appear distinguishable by their separations in the 0.56 to 0.66 µm interval.



Continuum-removal diagram showing the difference in relative reflectances for various vegetation types.

The technique can work particularly well in picking diagnostic bands for minerals that are very similar in crystal structure but differ in substitution of one chemical element (usually as an ion) for another. The common minerals, Calcite (CaCO3) and Dolomite (Ca,MgCO3) have a prominent absorption band near 2.3 µm, which reaches about the same depth in spectra of each species. The continuum-removal diagram for both shows that Dolomite reaches its trough point at a slightly lower wavelength.

This continuum-removal diagram shows that calcite (calcium carbonate) can be distinguished from dolomite (calcium-magnesium carbonate) by the slight difference in the wavelength of the low point of a major absorption band in the 2.3 mm part of the spectrum.

This approach to absorption band analysis has proved to be a powerful tool for enhancement and separation of small but often significant differences that allow us to properly identify materials (those belonging to related groups and those unrelated but with absorption bands that tend to coincide).