FREC 480 -- Intro to Remote Sensing

This page provides background information on remote sensing, including an overview of sensor technologies an a general discussion of image enhancement, registration and interpretation methods.  Subsequent pages detail the specifics of ArcView's Image Analyst extension.

Brief history of remote sensing

Remote sensing technologies have largely been driven by military intelligence needs.  Airphotos have been taken from balloons since the 19th century, and from airplanes since the early 20th century.  The US relied increasingly heavily on aerial surveillance during World War II.  During the Cold War the US used spy planes to monitor Soviet military installations, but by the late 1950's the Soviets had developed the ability to shoot these planes down. Surveillance airphotos of Cuba identified the presence of Soviet military advisors (a US analyst first noticed new soccer fields, and knew that Cubans play baseball but Russians play soccer), and led to detection of Soviet nuclear weapons in Cuba, triggering the Cuban Missile Crisis in 1962.

The need for more secure remote sensing platforms drove US development of unmanned systems, including satellite-based systems.  The early unmanned systems, involving rocket launches of robotic cameras that jettisoned film canisters, had very high rates of launch failure and canister recovery failure.  By the late 1950's the US had developed scanner technologies as a substitute for photography, and scanners mounted on earth satellites radioed images back to receiving stations in digital form, obviating the need for film canister recovery.

The electromagnetic spectrum

The electromagnetic spectrum includes visible light, radio waves, heat, X-rays and other forms of energy.  Electromagnetic radiation (EMR) travels at the speed of light (c = 3 x 108 meters/second) in a sinusoidal (wavy) pattern.  Wave components in the electric field and the magnetic field have defined wavelength and frequency.  The wavelength is the distance between wave peaks; the frequency is the number of peaks passing a fixed point per unit time.  The basic physical relationship between these is

c = wavelength x frequency
and since c is constant, wavelength and frequency are inversely related, so we can characterize EMR by either measure.  We typically classify EMR by wavelength, usually measured in micrometers (1 µm = 10-6m).  The electromagnetic spectrum includes cosmic rays (<10-7 µm), gamma rays (~10-6 µm), X-rays (~10-4 µm), ultraviolet (0.1--0.4 µm), visible light (0.4--0.7 µm), near infrared (0.7--1.3 µm), mid-infrared: (1.3--3 µm), thermal infrared (3--10 µm), microwave (~105 µm) and TV/radio (~108 µm).

Before reaching a remote sensor, EMR passes through the atmosphere, and is attenuated by atmospheric scattering and absorption.  Rayleigh scattering is caused by atmospheric molecules much smaller than the wavelengths of light passing through; the most familiar consequence of this is the blue sky, caused by selective Rayleigh scattering of more short-wavelength (UV and blue) light than long-wavelength (green and red) light.  Mie scattering is caused water vapor and dust.particles of similar size to the wavelengths passing through, and is generally less selective than Rayleight scattering.

Transmittance of light throught the Earth's atmosphere is strongly dependent on wavelength.  The atmosphere absorbs (blocks) most UV and various ranges of IR wavelengths between 1 and 10 µm.  It is most transparent to visible light (0.4--0.7 µm) and microwave and radio/TV wavelengths (>104 µm).  Sensors (including our eyes) are thus designed to exploit atmospheric transmissivity windows.

Healthy vegetation absorbs red wavelengths for photosynthesis, and reflects a high proportion of green light, so it appear green to our eyes.  But vegetation also reflects a very high proportion of infrared up to 1.3 µm, and has water absorption bands at 1.4, 1.9 and 2.7 µm--wavelengths that penetrate the leaf surface and are absorbed by interior water cells.  If our eyes were evolved to see near infrared, most vegetation would appear as bright infrared rather than green.  In fact, we can distinguish vegetation types more easily in the infrared than in the visible light spectrum, and can distinguish healthy from drought-stressed vegetation by the degree of IR absorption in the water absorption bands.

Clear water absorbs relatively little light below 0.6 µm, but is highly absorptive of near infrared.  In infrared band images open water features appear black, and moist soils appear darker than dry soils.  Water with excessive algae has reduced blue and increased green reflectance; water with suspended sediments has increased red reflectance.

A multispectral sensor samples EMR in various discrete wavelength ranges.  A panchromatic sensor samples EMR over a single broad range of wavelengths.  Multispectral imagery is useful for analyzing spectral signatures (patterns of reflectance in various wavelengths) of Earth features.  However, spectral signatures are not necessarily consistent through time or space.  For example, the spectral signature of corn will vary by soil type, rainfall, stage in the growing season, even sun angle and sensor angle.  These variations generally complicate change detection.

Airphoto technology

A photograph is typically printed from a negative generated by chemical reaction of an emulsion of silver halide crystals to light.  Panchromatic photography records exposures across a fairly broad range of wavelengths; for example, conventional black-and-white photography records light in the 0.4--0.7 µm visible light range on a single-layer emulsion of silver halide crystals.

Color photography involves three layers of emulsions of crystals that are sensitive to more specific wavelengths.  Typical color film has a blue-sensitive layer on the surface that forms yellow dye, underlain by a blue-blocking (yellow) filter; then a green- (and blue-) sensitive layer that forms magenta dye, then a red- (and blue-) sensitive layer that forms cyan dye.

Color infrared film substitutes an IR- (and blue-) sensitive top layer that forms cyan dye for the blue-sensitive top layer; the green-sensitive layer form yellow dye and the red-sensitive layer forms magenta dye.  The resulting print shows color-IR as red, red as green, and green as blue.  Color-IR file is used with a blue-absorbing filter.

Photographic resolution depends on the size of the silver halide grains in the emulsion.  Film with small grains is slower to react to light, and thus requires a slower shutter speed and/or a wider lens aperture which reduces depth of focus.  Film with large grains reacts to light faster, and thus permits a faster shutter speed and/or smaller lens aperture, but the grainer images are less suitable for enlargement.

Airphotos are conventionally taken with a large single-lens fixed focal-length camera mounted over a hatch in the underside of the airplane's fuselage.  The lens is designed to impart very low geometric distortion to the image.  The focal length (the distance between lens center and film, which is also the minimum distance from the lens at which light from an (approximately) infinite distance is focused on the film) ranges from 90 mm for low-altitude photography through 300 mm for very high altitude photography, with 152 mm being most common.  The lens is triggered at set intervals by an intervalometer as the pilot maintains a steady speed and altitude with no rolling over an assigned ground track.  Some cameras can compensate for blurring due to forward movement of the airplane by slowly moving the film across the focal plane in time with the image movement.  Strip cameras keep the lens shutter open continuously while the film is rolled in this manner, and yield continuous image strips.  Digital imaging technologies are gradually displacing photography.

All airphotos are perspective projections rather than idealized orthographic projections, although they can be satisfactory approximations of orthographic projections if taken from sufficient altitude.  The distortions of perspective can be problematic when the width of the scene or the variation in ground elevations is a significant fraction of the airplane height above the ground.  The right panel of this image illustrates an extreme case where relative positions, sizes, and feature footprint shapes are all grossly distorted.

Since geometric distortion of the image is greatest at image edges, parallel ground tracks are typically planned to yield 30 percent or more overlap on each side of the image.  The intervalometer timing yields 60 percent or more overlap on successive images of a ground track.  Multiple images can be tiled or mosaicked to create a continuous picture of the terrain.

Large overlaps permit stereo-imaging.  A stereo image is two images of the same subject taken from different angles; viewed through a stereoscope, the stereo image provides a sense of relative elevations of ground features via the relative displacements of features in the images.  In case you're not familiar with stereo images, here's a simple example of a stereogram (which probably only works on high-resolution monitors).  Staring straight through the image so you see a 3D image of an old-fashioned hand-held stereoscope with the words "STEREO SCOPE" floating above it.

The degree of vertical exaggeration in stereo images is a function of distances of feature centers versus image centers, divided by the distance from the image plane to the eye.

Airphotos: National High Altitude Photography (NHAP) and National Aerial Photography Program (NAPP)

The NHAP program was begun in 1980 under USGS to obtain airphoto coverage of the coterminous US every 5 years.  In 1987 the program was renamed NAPP.  NHAP airphotos were acquired at 40,000 altitude; flight lines were centered on the 1:24,000-scale USGS 7.5-minute quads. NHAP used two different camera systems: a 6 inch focal length lens for black-and-white film at an approximate scale of 1:80,000 and an 8.25 inch focal length lens was for color-infrared film at an approximate scale of 1:58,000.   NAPP photography are acquired at 20,000 feet altitude with a 6 inch focal length lens. Flight lines are quarter quad-centered on the 1:24,000-scale USGS maps. NAPP photographs have an approximate scale of 1:40,000, and are greyscale or color-IR.  Spatial resolution is typically 1-2 meters.

USGS has completed two NAPP cycles, and now doing a 7-year third cycle (1997-2003) via cost-sharing agrements with other federal agencies and state governments.  The NHAP/NAPP photo archive is huge.   You can acquire film negatives or positives or prints in various sizes and scales, then scan these and register them with other GIS data.  See the NAPP section of USGS's Global Land Information System (GLIS) for info and on-line ordering.

Digital Orhtophotos

A digital orthophoto is an airphoto that has been scanned and registered to geographic coordinates.  Delaware has two recent series of digital orthophotos, created by EarthData, Inc. under contract with the State of Delaware.  Each orthophoto covers one quarter of a 7.5-minute quad at a ground resolution of 1 meter.  Delaware's 1992 series are color-IR in SPOT Image (.BIP/.BIL) format; Delaware's 1997 series are grayscale in .TIFF format.  These orthophoto series are the base for two Land-Use/Land Cover mappings that have been done for Delaware.

5-meter resolution versions of both series are available off the Spatial Analysis Lab web site in JPEG format.  To use these in ArcView, also download the companion world file with each image.  The lab also distributes 1-meter resolution JPEG versions of these orthophotos, 1 CD per county.

Scanner technology

Direct scanning of terrain using arrays of charged-couple devices (CCD's) offered various advantages over photography.  While photography is generally limited to a spectral range of 0.3--0.9 µm (UV, visible and near-IR), multispectral scanners have a potential range of 0.3--14 µm (UV and visible, IR and thermal IR).  Scanners are somewhat easier to calibrate radiometrically, and they don't run out of film.  A scan yields a regular grid array of reflectance values; in contrast to a photograph's somewhat irregular grain sizes and random placements.

There are two basic scanner designs for remote sensing.  A "whisk-broom" scanner uses an oscillating mirror to scan terrain reflectances along scan lines perpendicular to the sensor's flight line, reflecting the light to a single fixed sensor element, or through a prism to an array of fixed sensor elements recording different wavelengths.  The sensor quantizes sampled analog reflectances to digital numbers (DN's, typically 0-255) at set time intervals.  The mirror oscillation is fast enough that the scan line strips are contiguous or slightly overlapping.

The second design is a "push-broom" scanner that uses a linear array of sensor elements to record reflectances at uniform intervals along the scan line simultaneously.  The sensor elements quantize sampled reflectances to DN's at set time intervals so that the scan lines are contiguous or slightly overlapping.  The sampling rate is typically at least twice the highest frequency in the signal.

Landsat

Building on the successes of weather satellites and photography from space capsules in the early 1960's, NASA begain designing a series of satellite platforms for Earth imaging in 1967, and launched Landsat 1 in July, 1972.  The first five Landsat satellites carried Multi-Spectral Scanner (MSS) sensors and Return Beam Vidicon (RBV) cameras, as well as solar arrays to maintain power and antennae for data transmission.  The first three Landsats were placed in near-polar orbits about 900km above the Earth, with orbital periods of about 103 minutes.  The orbits are designed so that the satellite is on a southbound track on the sunlit side of the Earth, and crosses the Equator at the same local sun time (about 9:40AM) on every orbit.  The sun-synchronous orbit is intended to maximize the uniformity of the sun angle on each track, although it cannot compensate for seasonal or latitudinal changes in sun angle.  Successive satellite tracks are 2,760 km apart at the equator.  The satellite travels about 25,000 km/hour.

Landsat MSS sensors use a whisk-broom design, and records reflectance intensities in four wavelength bands: 0.5.-0.6 µm (green), 0.6-0.7 µm (red), 0.7-0.8 µm (near-IR) and 0.8-1.1 µm (IR).   Each west-to-east sweep of the oscillating mirror samples six contiguous 185 km lines of 79x79-meter pixels--nominally 3,240 pixels per swath.  These are recorded by six arrays of four sensor, one sensor per wavelength band.  The brightness values are then converted to 6-bit DN's (0-63).  A standard MSS scene is 185 x 185 km: with about 2,340 scan lines of 3,240 pixels per line, so each of the four bands in a scene has about 7.6 million pixels, all recorded in about 27 seconds.  The data generation rate is 15 MB/second.  Landsat scenes are referenced by orbital path (numbered 0-251 east to west) and row (numbered from north to south, with row 60 at the equator).  Given the 185-km path width, the satellite revisits each path every 18 days.

The first three Landsats were decomissioned by March, 1983.  Landsats 4 and 5 were launched in July, 1982, and March, 1984, and carry Thematic Mapper (TM) sensors as well as MSS sensors.  These satellites use a somewhat lower-altitude orbit (705 km) and thus travel slightly faster; they revisit the same path every 16 days.  The TM contains 7 sensor types:
 
Band
Wavelength
Res. 
Characteristics; Applications
1: blue
0.45-0.52 µm
30m
best water penetration; coastal mapping, vegetation, cultural features
2: green
0.52-0.60 µm
30m
reflected by vegetation; vegetation, cultural features
3: red
0.63-0.69 µm
30m
chlorophyll absorption region; vegetation, cultural features
4: near-IR
0.76-0.90 µm
30m
reflected by vegetation, abosrbed by water; vegetation, soil moisture assessment
5: mid-IR
1.55-1.75 µm
30m
absorbed by water; moisture
6: thermal IR
10.4-12.5 µm
120m
120 m resolution, operational on night side of orbit also; thermal mapping
7: mid-IR
2.08-2.35 µm
30m
largely redundant with band 5; geologic classification

Here is a set of TM band files from July 30, 1988, covering Newark, DE.  Each band has been histogram-equalized for clarity.  Note the relative blur in thermal band 6, and the similarities between mid-IR bands 5 and 7.

TM has a total of 100 individual sensors; each mirror oscillation scans 16 contiguous 30-m-wide lines in each of 6 non-thermal bands, plus 4 120-m-wide thermal lines.  TM's data generation rate is 85 MB/second.

Landsat 6, carrying an Enhanced Thematic Mapper (ETM), failed upon launch in 1993.  Landsat 7 (shown in the NASA diagram above) was launched in April, 1999, and its ETM+ scanner contains eight sensor types: five sensors sample visible and near-IR reflectances (bands 1,2,3,4 and 8 panchromatic) between 0.4 and 1.0 µm; two sensors sample mid-IR (bands 5 and 7) between 1.0 and 3.0 µm; and one sensor samples thermal IR (band 6) at 8.0-12.0 µm.  All bands are sampled at 30 m nominal ground resolution except thermal band 6 (60 m) and panchromatic band 8 (15 m).  Like TM, ETM+ scans 16 simultaneous lines per non-thermal band (32 panchromatic lines).  Landsat 7 has similar orbit characterisics to Landsats 4 and 5--a 705 km orbit elevation and a 16-day path revisit interval.  ETM+ has a data generation rate of 150MB/second.

SPOT

SPOT (Systeme Pour l'Observation de la Terre) is a European satellite consortium established by the French government in 1978.  The company launched SPOT-1 in 1986 (on stand-by service since 1990), SPOT-2 in 1990 (still operational), SPOT-3 in 1993 (failed in 1997) and SPOT-4 in 1998.  SPOT satellites also use near-polar sun-synchronous orbits at an altitude of 832 km, crossing the equator at 10:30 AM local sun time.

Each SPOT satellite carries twin High Resolution Visible (HRV) scanners.  Each includes a 10-meter resolution panchromatic band (0.51 to 0.73 µm) and three 20-meter resolution multi-spectral bands: green (0.50-0.59 µm), red (0.61--0.68 µm) and near-IR (0.79--0.89 µm).  SPOT-4's scanners also include 10-meter resolution mono-spectral sensors (0.61-0.68 µm) and 20-meter resolution mid-IR sensors (1.58-1.75 µm).  HRV scanners use a push-broom design with a 3,000-element linear arrays of CCD's for each multispectral band and a 6,000-element array for the panchromatic band.  A tilting mirror that allows each scanner to image to the left or right of the ground track.  The imaging swath at nadir (straight down) is 60 km wide; at maximum mirror tilt the scanner images a swath of 80 km at a 27-degree angle to the left or right.  Although the path revisit rate at nadir is 26 days, off-nadir imaging capabilities permits imaging the same scene at intervals of one to five days.  Off-nadir imaging also supports stereoscopic imaging.  While the tilting abilities of the HRV scanner increase its flexibility, they can also require more complex geometric correction of images.

AVHRR

NOAA's Advanced Very High Resolution Radiometers (AVHRR) are 4- or 5-band multispectral scanners, sampling in the red (0.58-0.68 µm), near-IR (0.725-1.10 µm), mid-IR (3.55-3.93 µm) and thermal IR (10.50-11.50 µm and, on some satellites, 11.5-12.5 µm).  AVHRR satellite orbit altitude is 833 km; the AVHRR sensor images a 2399 km wide swath, and thus revisits all of the Earth daily.  Image resolution is 1.1 x 4 km (the satellite performs on-board sampling, averaging successive scan lines and reporting every third line).

CORONA

Image archives from the US military's CORONA satellite system were declassified in 1995.  The CORONA program begain in 1959.  Imagery are available for the years 1960 through 1972.  Spatial resolutions vary; the best resolution is about 2x2 meters.
An automated index of the entire film collection is accessible through the USGS Global Land Information System (GLIS).

IKONOS

The IKONOS satellite is one of the new generation of Earth imaging satellites.  It was launched on September 24, 1999 and the first images of Washington, DC (see below), were released October 12.  Orbit frequency is 98 minutes at an altitude of 681 km; orbit is near-polar and sun-synchronous, crossing the equator at 10:30 AM local sun time.  IKONOS carries a 1-meter resolution panchromatic (0.4 -0.90 µm) scanner and a 4-meter resolution 4-band multispectral scanner recording the same wavelengths as TM bands 1-4.  The sensor is tiltable, and has a 13-km swath width at nadir.  Standard scenes are 13 x 13 km.  Imagery is distributed by Space Imaging, Inc.

Ordering satellite imagery

You can order MSS images and TM images that are at least 10 years old from USGS's EROS Data Center.  For example, the Newark images above were extracted from a full image comprised of two consecutive scenes from rows 32-34 of path 14.  The cost was $625 plus a few dollars for shipping (EDC's charges $425 for one scene, $200 extra for the second).  EDC's web-site lets you preview and order scenes on-line.  I framed the Delmarva peninsula by rolling the scenes ahead by 60%, so this scene contains the bottom 40% of row 32, all of row 33 and the top 60% of row 34.  EDC will also perform standard radiometric corrections and rough registration of imagery to any standard projection you specify.  The image shown here is a composite of bands 3, 2 and 1 simulating a conventional color photo.  Water areas were identified as very low values in band 4, and masked out of this image.  Landsat 7 ETM+ imagery is also available through EDC at a cost of $600/scene.  TM imagery less than 10 years old can be ordered from Space Imaging (formerly EOSAT) at substantially higher cost.

You can order SPOT imagery in various states (basic radiometric correction only through fully registered) from SPOT Image Corporation.  SPOT frequently offers discounts to educational institutions.

You can order AVHRR imagery from USGS's EROS Data Center; typical costs are $50/scene.