When radiation interacts with matter, some wavelengths are absorbed and others are reflected.To enhance features in image data, it is necessary to understand how vegetation, soils, water, and other land covers reflect and absorb radiation. The study of the absorption and reflection of EMR waves is called spectroscopy.
Spectroscopy
Most commercial sensors, with the exception of imaging radar sensors, are passive solar imaging sensors. Passive solar imaging sensors can only receive radiation waves; they cannot transmit radiation. (Imaging radar sensors are active sensors that emit a burst of microwave radiation and receive the backscattered radiation.)
The use of passive solar imaging sensors to characterize or identify a material of interest is based on the principles of spectroscopy.
Therefore, to fully utilize a visible/infrared (VIS/IR) multispectral data set and properly apply enhancement algorithms, it is necessary to
understand these basic principles. Spectroscopy reveals the:
• absorption spectra—the EMR wavelengths that are absorbed by specific materials of interest
• reflection spectra—the EMR wavelengths that are reflected by specific materials of interest
Absorption Spectra
Absorption is based on the molecular bonds in the (surface) material.
Which wavelengths are absorbed depends upon the chemical composition and crystalline structure of the material. For pure compounds, these absorption bands are so specific that the SWIR region is often called an infrared fingerprint.
Atmospheric Absorption
In remote sensing, the sun is the radiation source for passive sensors.
However, the sun does not emit the same amount of radiation at all wavelengths. Figure 4 shows the solar irradiation curve, which is far from linear.
Figure 4: Sun Illumination Spectral Irradiance at the Earth’s Surface
Source: Modified from Chahine et al, 1983
Solar radiation must travel through the Earth’s atmosphere before it reaches the Earth’s surface. As it travels through the atmosphere, radiation is affected by four phenomena (Elachi, 1987):
• absorption—the amount of radiation absorbed by the atmosphere
• scattering—the amount of radiation scattered away from the field of view by the atmosphere
• scattering source—divergent solar irradiation scattered into the field of view
• emission source—radiation re-emitted after absorption 0
0.0 Eλ Spectral Irradiance (Wm-2 m-1 )
3.0
1.5 1.8 2.1 2.4 2.7
1.2 0.9
0.6 0.3
Wavelength μm 2500
2000
1500
1000
500
UV VIS INFRARED
Solar irradiation curve outside atmosphere
Solar irradiation curve at sea level
Peaks show absorption by H20, C02, and O3
Figure 5: Factors Affecting Radiation
Source: Elachi, 1987
Absorption is not a linear phenomena—it is logarithmic with concentration (Flaschka, 1969). In addition, the concentration of atmospheric gases, especially water vapor, is variable. The other major gases of importance are carbon dioxide (CO2) and ozone (O3), which can vary considerably around urban areas. Thus, the extent of
atmospheric absorbance varies with humidity, elevation, proximity to (or downwind of) urban smog, and other factors.
Scattering is modeled as Rayleigh scattering with a commonly used algorithm that accounts for the scattering of short wavelength energy by the gas molecules in the atmosphere (Pratt, 1991)—for example, ozone. Scattering is variable with both wavelength and atmospheric aerosols. Aerosols differ regionally (ocean vs. desert) and daily (for example, Los Angeles smog has different concentrations daily).
Scattering source and emission source may account for only 5% of the variance. These factors are minor, but they must be considered for accurate calculation. After interaction with the target material, the reflected radiation must travel back through the atmosphere and be subjected to these phenomena a second time to arrive at the satellite.
Absorption—the amount of
Scattering—the amount of radiation
Scattering Source—divergent solar
Emission Source—radiation
Radiation
radiation absorbed by the atmosphere
re-emitted after absorption
scattered away from the field of view
irradiations scattered into the field of view
by the atmosphere
The mathematical models that attempt to quantify the total atmospheric effect on the solar illumination are called radiative transfer equations.
Some of the most commonly used are Lowtran (Kneizys et al, 1988) and Modtran (Berk et al, 1989).
See "Enhancement" on page 455 for more information on atmospheric modeling.
Reflectance Spectra
After rigorously defining the incident radiation (solar irradiation at target), it is possible to study the interaction of the radiation with the target material. When an electromagnetic wave (solar illumination in this case) strikes a target surface, three interactions are possible (Elachi, 1987):
• reflection
• transmission
• scattering
It is the reflected radiation, generally modeled as bidirectional
reflectance (Clark and Roush, 1984), that is measured by the remote sensor.
Remotely sensed data are made up of reflectance values. The resulting reflectance values translate into discrete digital numbers (or values) recorded by the sensing device. These gray scale values fit within a certain bit range (such as 0 to 255, which is 8-bit data) depending on the characteristics of the sensor.
Each satellite sensor detector is designed to record a specific portion of the electromagnetic spectrum. For example, Landsat Thematic Mapper (TM) band 1 records the 0.45 to 0.52 μm portion of the spectrum and is designed for water body penetration, making it useful for coastal water mapping. It is also useful for soil/vegetation discriminations, forest type mapping, and cultural features identification (Lillesand and Kiefer, 1987).
The characteristics of each sensor provide the first level of constraints on how to approach the task of enhancing specific features, such as vegetation or urban areas. Therefore, when choosing an enhancement technique, one should pay close attention to the characteristics of the land cover types within the constraints imposed by the individual sensors.
The use of VIS/IR imagery for target discrimination, whether the target is mineral, vegetation, man-made, or even the atmosphere itself, is based on the reflectance spectrum of the material of interest (see Figure 6). Every material has a characteristic spectrum based on the chemical composition of the material. When sunlight (the illumination source for VIS/IR imagery) strikes a target, certain wavelengths are absorbed by the chemical bonds; the rest are reflected back to the sensor. It is, in fact, the wavelengths that are not returned to the sensor that provide information about the imaged area.
Specific wavelengths are also absorbed by gases in the atmosphere (H2O vapor, CO2, O2, and so forth). If the atmosphere absorbs a large percentage of the radiation, it becomes difficult or impossible to use that particular wavelength(s) to study the Earth. For the present Landsat and Systeme Pour l’observation de la Terre (SPOT) sensors, only the water vapor bands are considered strong enough to exclude the use of their spectral absorption region. Figure 6 shows how Landsat TM bands 5 and 7 were carefully placed to avoid these regions. Absorption by other atmospheric gases was not extensive enough to eliminate the use of the spectral region for present day broad band sensors.
Figure 6: Reflectance Spectra
100
80
60
40
20
0
Reflectance, %
.4 .6 .8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 Wavelength, μm
Vegetation (green)
Silt loam
Atmospheric bands
4 5 6 7 Landsat MSS bands
1 2 3 4 5 7
Landsat TM bands
Kaolinite
absorption
Source: Modified from Fraser, 1986;Crist et al, 1986; Sabins, 1987 NOTE: This chart is for comparison purposes only. It is not meant to show actual values. The spectra are offset to better display the lines.
An inspection of the spectra reveals the theoretical basis of some of the indices in the ERDAS IMAGINE Image Interpreter. Consider the vegetation index TM4/TM3. It is readily apparent that for vegetation this value could be very large. For soils, the value could be much smaller, and for clay minerals, the value could be near zero. Conversely, when the clay ratio TM5/TM7 is considered, the opposite applies.
Hyperspectral Data
As remote sensing moves toward the use of more and narrower bands (for example, AVIRIS with 224 bands each only 10 nm wide),
absorption by specific atmospheric gases must be considered. These multiband sensors are called hyperspectral sensors. As more and more of the incident radiation is absorbed by the atmosphere, the digital number (DN) values of that band get lower, eventually becoming useless—unless one is studying the atmosphere. Someone wanting to measure the atmospheric content of a specific gas could utilize the bands of specific absorption.
NOTE: Hyperspectral bands are generally measured in nanometers (nm).
Figure 6 shows the spectral bandwidths of the channels for the Landsat sensors plotted above the absorption spectra of some common natural materials (kaolin clay, silty loam soil, and green vegetation). Note that while the spectra are continuous, the Landsat channels are segmented or discontinuous. We can still use the spectra in interpreting the Landsat data. For example, a Normalized Difference Vegetation Index (NDVI) ratio for the three would be very different and, therefore, could be used to discriminate between the three materials. Similarly, the ratio TM5/TM7 is commonly used to measure the concentration of clay minerals. Evaluation of the spectra shows why.
Figure 7 shows detail of the absorption spectra of three clay minerals.
Because of the wide bandpass (2080 to 2350 nm) of TM band 7, it is not possible to discern between these three minerals with the Landsat sensor. As mentioned, the AVIRIS hyperspectral sensor has a large number of approximately 10 nm wide bands. With the proper selection of band ratios, mineral identification becomes possible. With this data set, it would be possible to discriminate between these three clay minerals, again using band ratios. For example, a color composite image prepared from RGB = 2160nm/2190nm, 2220nm/2250nm, 2350nm/2488nm could produce a color-coded clay mineral image-map.
The commercial airborne multispectral scanners are used in a similar fashion. The Airborne Imaging Spectrometer from the Geophysical &
Environmental Research Corp. (GER) has 79 bands in the UV, visible, SWIR, and thermal-infrared regions. The Airborne Multispectral Scanner Mk2 by Geoscan Pty, Ltd., has up to 52 bands in the visible, SWIR, and thermal-infrared regions. To properly utilize these
hyperspectral sensors, you must understand the phenomenon involved and have some idea of the target materials being sought.
Figure 7: Laboratory Spectra of Clay Minerals in the Infrared Region
Source: Modified from Sabins, 1987
NOTE: Spectra are offset vertically for clarity.
2000 2200 2400 2600
Landsat TM band 7
2080 nm 2350 nm
Kaolinite
Montmorillonite
Illite
Reflectance, %
Wavelength, nm
The characteristics of Landsat, AVIRIS, and other data types are discussed in "Raster and Vector Data Sources" on page 55. See
"Enhancement" on page 455 for more information on the NDVI ratio.
Imaging Radar Data
Radar remote sensors can be broken into two broad categories:
passive and active. The passive sensors record the very low intensity, microwave radiation naturally emitted by the Earth. Because of the very low intensity, these images have low spatial resolution (that is, large pixel size).
It is the active sensors, termed imaging radar, that are introducing a new generation of satellite imagery to remote sensing. To produce an image, these satellites emit a directed beam of microwave energy at the target, and then collect the backscattered (reflected) radiation from the target scene. Because they must emit a powerful burst of energy, these satellites require large solar collectors and storage batteries. For this reason, they cannot operate continuously; some satellites are limited to 10 minutes of operation per hour.
The microwave energy emitted by an active radar sensor is coherent and defined by a narrow bandwidth. The following table summarizes the bandwidths used in remote sensing.
*Wavelengths commonly used in imaging radars are shown in parentheses.
Band Designation* Wavelength (λ), cm Frequency (υ), GHz (109 cycles · sec-1)
Ka (0.86 cm) 0.8 to 1.1 40.0 to 26.5
K 1.1 to 1.7 26.5 to 18.0
Ku 1.7 to 2.4 18.0 to 12.5
X (3.0 cm, 3.2 cm) 2.4 to 3.8 12.5 to 8.0
C 3.8 to 7.5 8.0 to 4.0
S 7.5 to 15.0 4.0 to 2.0
L (23.5 cm, 25.0 cm) 15.0 to 30.0 2.0 to 1.0
P 30.0 to 100.0 1.0 to 0.3
A key element of a radar sensor is the antenna. For a given position in space, the resolution of the resultant image is a function of the antenna size. This is termed a real-aperture radar (RAR). At some point, it becomes impossible to make a large enough antenna to create the desired spatial resolution. To get around this problem, processing techniques have been developed which combine the signals received by the sensor as it travels over the target. Thus, the antenna is perceived to be as long as the sensor path during backscatter reception. This is termed a synthetic aperture and the sensor a synthetic aperture radar (SAR).
The received signal is termed a phase history or echo hologram. It contains a time history of the radar signal over all the targets in the scene, and is itself a low resolution RAR image. In order to produce a high resolution image, this phase history is processed through a hardware/software system called an SAR processor. The SAR processor software requires operator input parameters, such as information about the sensor flight path and the radar sensor's characteristics, to process the raw signal data into an image. These input parameters depend on the desired result or intended application of the output imagery.
One of the most valuable advantages of imaging radar is that it creates images from its own energy source and therefore is not dependent on sunlight. Thus one can record uniform imagery any time of the day or night. In addition, the microwave frequencies at which imaging radars operate are largely unaffected by the atmosphere. This allows image collection through cloud cover or rain storms. However, the
backscattered signal can be affected. Radar images collected during heavy rainfall are often seriously attenuated, which decreases the signal-to-noise ratio (SNR). In addition, the atmosphere does cause perturbations in the signal phase, which decreases resolution of output products, such as the SAR image or generated DEMs.
Resolution
Resolution is a broad term commonly used to describe:• the number of pixels you can display on a display device, or
• the area on the ground that a pixel represents in an image file.
These broad definitions are inadequate when describing remotely sensed data. Four distinct types of resolution must be considered:
• spectral—the specific wavelength intervals that a sensor can record
• spatial—the area on the ground represented by each pixel
• radiometric—the number of possible data file values in each band (indicated by the number of bits into which the recorded energy is divided)
• temporal—how often a sensor obtains imagery of a particular area These four domains contain separate information that can be extracted from the raw data.
Spectral Resolution
Spectral resolution refers to the specific wavelength intervals in the electromagnetic spectrum that a sensor can record (Simonett et al, 1983). For example, band 1 of the Landsat TM sensor records energy between 0.45 and 0.52 μm in the visible part of the spectrum.Wide intervals in the electromagnetic spectrum are referred to as coarse spectral resolution, and narrow intervals are referred to as fine spectral resolution. For example, the SPOT panchromatic sensor is considered to have coarse spectral resolution because it records EMR between 0.51 and 0.73 μm. On the other hand, band 3 of the Landsat TM sensor has fine spectral resolution because it records EMR between 0.63 and 0.69 μm (Jensen, 1996).
NOTE: The spectral resolution does not indicate how many levels the signal is broken into.
Spatial Resolution
Spatial resolution is a measure of the smallest object that can be resolved by the sensor, or the area on the ground represented by each pixel (Simonett et al, 1983). The finer the resolution, the lower the number. For instance, a spatial resolution of 79 meters is coarser than a spatial resolution of 10 meters.Scale
The terms large-scale imagery and small-scale imagery often refer to spatial resolution. Scale is the ratio of distance on a map as related to the true distance on the ground (Star and Estes, 1990).
Large-scale in remote sensing refers to imagery in which each pixel represents a small area on the ground, such as SPOT data, with a spatial resolution of 10 m or 20 m. Small scale refers to imagery in which each pixel represents a large area on the ground, such as Advanced Very High Resolution Radiometer (AVHRR) data, with a spatial resolution of 1.1 km.
This terminology is derived from the fraction used to represent the scale of the map, such as 1:50,000. Small-scale imagery is represented by a small fraction (one over a very large number). Large-scale imagery is represented by a larger fraction (one over a smaller number).
Generally, anything smaller than 1:250,000 is considered small-scale imagery.
NOTE: Scale and spatial resolution are not always the same thing. An image always has the same spatial resolution, but it can be presented at different scales (Simonett et al, 1983).
Instantaneous Field of View
Spatial resolution is also described as the instantaneous field of view (IFOV) of the sensor, although the IFOV is not always the same as the area represented by each pixel. The IFOV is a measure of the area viewed by a single detector in a given instant in time (Star and Estes, 1990). For example, Landsat MSS data have an IFOV of 79 × 79 meters, but there is an overlap of 11.5 meters in each pass of the scanner, so the actual area represented by each pixel is 56.5 × 79 meters (usually rounded to 57 × 79 meters).
Even though the IFOV is not the same as the spatial resolution, it is important to know the number of pixels into which the total field of view for the image is broken. Objects smaller than the stated pixel size may still be detectable in the image if they contrast with the background, such as roads, drainage patterns, and so forth.
On the other hand, objects the same size as the stated pixel size (or larger) may not be detectable if there are brighter or more dominant objects nearby. In Figure 8, a house sits in the middle of four pixels. If the house has a reflectance similar to its surroundings, the data file values for each of these pixels reflect the area around the house, not the house itself, since the house does not dominate any one of the four pixels. However, if the house has a significantly different reflectance than its surroundings, it may still be detectable.
Figure 8: IFOV
Radiometric Resolution
Radiometric resolution refers to the dynamic range, or number of possible data file values in each band. This is referred to by the number of bits into which the recorded energy is divided.20 m
20 m
house 20 m
20 m
For instance, in 8-bit data, the data file values range from 0 to 255 for each pixel, but in 7-bit data, the data file values for each pixel range from 0 to 128.
In Figure 9, 8-bit and 7-bit data are illustrated. The sensor measures the EMR in its range. The total intensity of the energy from 0 to the maximum amount the sensor measures is broken down into 256 brightness values for 8-bit data, and 128 brightness values for 7-bit data.
Figure 9: Brightness Values
Temporal Resolution
Temporal resolution refers to how often a sensor obtains imagery of a particular area. For example, the Landsat satellite can view the same area of the globe once every 16 days. SPOT, on the other hand, can revisit the same area every three days.NOTE: Temporal resolution is an important factor to consider in change detection studies.
Figure 10: illustrates all four types of resolution:
Figure 10: Landsat TM—Band 2 (Four Types of Resolution) 8-bit
Source: EOSAT
Data Correction
There are several types of errors that can be manifested in remotely sensed data. Among these are line dropout and striping. These errors can be corrected to an extent in GIS by radiometric and geometric correction functions.NOTE: Radiometric errors are usually already corrected in data from EOSAT or SPOT.
See "Enhancement" on page 455 for more information on radiometric and geometric correction.