The following is an overview of the subject of remote sensing. It was prepared by Nancy Gordon of the Centre for Environmental Applied Hydrology in the Department of Civil and Agricultural Engineering, and modified by Cliff Ogleby.
Remote sensing, a term coined in 1960, is the "measurement or acquisition of information by a recording device that is not in physical contact with the object under study". This definition includes more than just imagery, such as the remote acquisition of data on snow water contents or wildlife movements. In the environmental sciences, however, remote sensing is defined more precisely as the use of electromagnetic radiation sensors to record images of the environment, and their interpretation. In a sense, it includes conventional aerial photography and photo-interpretation, although these will be treated as a separate topic.
The amount of information which can be gained about an area is greatly enhanced by the acquisition of imagery. Imagery is attractive, too, where field survey costs are high or where regions are physically inaccessible. Environmental and resource surveys, crop conditions, snow cover, extent of urban area development or timber harvest, and the assessment of damage from bushfires, flooding or drought are some of the practical uses of remotely sensed images. Compared to a map, a photograph is much "closer to reality" Additionally, one is free to make one's own interpretations rather than relying on those of a mapmaker. The amount of information on an image may at first seem overwhelming because of this lack of prior interpretation, but with experience and repeated association with field conditions, qualities can be quickly inferred from imagery.
Maps, photographs and remotely sensed images each provide their own brand of information. The faces of planet earth can be viewed in total via high-altitude space imagery, whereas large-scale colour or black-and-white photos are more appropriate for detailed terrain studies. Knowledge of each product and its advantages and limitations can help in the selection of the best type of imagery for one's own specialized needs.
Remote sensing, particularly photo-interpretation, relies on the human cognitive system to extract information from an image. This process is often broken down to seven attributes that are considered when interpreting images, namely size, shape, shadow, tone texture, pattern and place. Combinations of these components of analysis facilitate the interpretation and recognition of objects in a scene, be it a photograph, television picture or for that matter, reality.
In the science of image interpretation, a distinction is made between an image and a photograph. A photograph is a chemical based recording of a scene, where the 'information' is carried by grains of silver halides suspended in a colloid. A normal 23cm aerial photograph has close to 7 billion grains of silver, all of which are capable of carrying information. An image is deemed to be an electronic recording of a scene, in particular one stored in a digital format. A satellite scene containing 3548 columns and 2983 lines contains around 10 million picture elements. Remote sensing systems in the context of this subject, generally create images.
Remotely sensed images can be divided into two types: data collected by passive systems which sense natural radiation, and data collected by active sensing systems in which electromagnetic radiation is emitted and the reflected signal detected by a sensor. Passive systems detect radiation emitted within a specific wavelength range such as visual or infrared, and are most often associated with satellites such as Landsat and SPOT. Active systems include sonar (sound navigation and ranging) and radar (radio detection and ranging). Examples of active and passive systems and their applications are given later.
A digital image consists of picture elements, called pixels in jargon. These pixels are effectively grid cells in a matrix, where the information associated with each is a measure of light reflectance. or intensity. Remotely sensed images generally have multiple bands, where each band contains pixels corresponding to a certain wavelength band of radiation.
Digital images are generally supplied in a computer compatible format, either on floppy disks for small scenes or on magnetic tape. A computer based image display and interpretation system is necessary in order to view the image. In the early days of image processing these systems needed large computers and specialized graphic workstations to perform these functions, now however it is possible to use PC type computers and common graphics display devices like VGA boards and screens. The cost of an image processing system has dropped from say $250,000 in the late 1970s to $8000 today, making the technology available to many different users in all countries around the world. Image processing and the subsequent information is no longer the sole right of technology rich nations.
The largest applications of remotely sensed data are for mapping land use and monitoring changes in characteristics of the Earth's surface. As compared to aerial photographs, satellite data tends to be less expensive per unit area covered As with photographs, satellite data are used for environmental and resource surveys - at up to a global scale.
The science of remote sensing was rapidly propelled into the future with the launching of the first Landsat satellite in the 1972. Since then, millions of images have been obtained for the evaluation of earth resources. The Landsat satellites carry two sensors: a multispectral scanning system (MSS) and either a Return Beam Vidicon (RBV) panchromatic television camera (which was of good image geometry but poor resolution, and called bands 1, 2 and 3), or, more recently, a Thematic Mapper (TM) scanner. Sensors record the intensity of reflected light of selected wavelength bands, and transmits this from the satellite to ground stations as digital numbers. Computers process the data to recreate a false colour composite image. Landsat satellites now repeat coverage of the same area every 16 days. The potential amount of data is mindboggling, and not surprisingly, not all possible images are collected or processed. The development of new products and software has grown rapidly in response to the need for efficient methods of filtering, error-checking, classifying and interpreting the digital data.
MSS data consists of four images of a scene taken in green,
red and two near-IR wavebands. Sensors detect levels of radiance,
which is recorded as a grey level value for each pixel,
the unit of resolution of an image. Ground resolution is 79
metres, although the pixel size is 56m by 79m, and is actually
trapezoidal in shape as a result of the sensor viewing angle and
earth rotation. Each image covers a ground area of 185 km by
185 km
 |  |
The Thematic Mapper, carried on Landsat satellites starting with Landsat 4 in 1982, record 256 radiance levels in 7 wavebands: blue/green, green, red, near-IR, near-middle IR, middle-IR and thermal IR. The pixel size is 30m by 30m for 6 of the 7 sensors; the thermal sensor has a lower resolution. Each image covers the same area as MSS data. Different properties of a surface can be ascertained from each wavelength or from combinations of wavelengths. For example, blue/green, green and near-IR are good for detecting vegetation reflectance and near-middle-IR and thermal IR are moisture sensitive. Thermal IR scanners, developed for "nocturnal snooping" by the military, have special applications related to temperature differences: location of frost hollows, estimation of plant moisture stress, location of thermal water pollution or hot springs, and the location and extent of vulcanism or above and below-ground wildfires.
SLAR, or "sideways-looking airborne radar" senses terrain to the side of an aircraft's path by sending out long (up to radio) wavelengths and recording the returned pulses Radar imagery is particularly useful for terrain analysis. It yields images with long "shadows" which enhances microtopography and relief; and is also sensitive to surface soil moisture. The major advantage of radar is that it has better than 99% cloud penetration ability, so is useful where cloud cover restricts the use of conventional photography or satellite data. The relationships between image tone and characteristics of the land surface are different than on conventional photography, so it will take time for SLAR imagery to gain acceptance. Although the cost is relatively high, SLAR imagery has potential in hydrological studies for detecting drainage patterns and mapping land cover, soil moisture and vegetation.
There is a vast amount of imagery available, although some governments place restrictions on the use of coverage within their territory. Landsat is operated in international public domain, and data are available free of access or copyright controls. Data is also available from earth-sensing satellites of other countries such as the SPOT (Système Probatoire dObservation de la Terre) satellite, developed by the French space center CNES. SPOT collects panchromatic data (with 10m spatial resolution) and MSS data (20-25m resolution) on a 26 day repeat cycle, dependent on the latitude and sensor angle.
In the United States, the EROS Data Center (Sioux Falls, South Dakota, 57198, USA) is the national centre for processing and disseminating not only Landsat data, but other spacecraft and aircraft-acquired imagery. EARTHNET is an organization within the European Space Agency which supplies a variety of satellite remote sensing data for Western Europe. Satellite data may be acquired as a false-colour "photograph" or on computer-compatible media as digital data.
A number of computer-assisted techniques are now available for rectifying an image, contrast stretching, colour and edge enhancement, filtering, and density slicing for analyzing an image based on groupings of image regions with the same range of grey levels. The same cautions are advised as for hydrologic data: know how the data was collected - whether from the same or different sensors or satellites; be aware of possible errors from noise in a signal, prevailing environmental conditions or variations in altitude, attitude and velocity of the satellite, and only push the information as far as its quality will allow.
The computer enhancement of digital images allows correction for errors in the data, the filtering of noise, the enhancement of texture and the clustering of pixels into 'classes'. Most of these operations involve the use of statistical processes on the digital numbers in the image. Shown below are some of the types of operations, and the effect they have on the displayed image.
 | |
|
| |
 |  |
|
|
|
Remote sensing is best practiced with knowledge of field conditions or the "ground truth" otherwise the patterns on a photograph or numbers on a magnetic tape will have little meaning. Although satellite data is increasingly treated by semi-automatic computer analysis, most operational uses of remote sensing still involve some human interpretive procedures. Interpretation of products in photographic format is more art than science, based on the ability of an interpreter to integrate the patterns on an image either consciously or subconsciously. It is a deductive process to go from qualitative information such as tone and/or color hue, location, shape, size, pattern, shadow, texture - and depth, from stereoscopic views - to the identification of relief features or vegetation types or cows.
As well as identification, conditions of resources can also be assessed by inference from what is observed on an image; for example, soil moisture conditions may be derived from the tone of vegetation. Experienced photo-interpreters work from familiar features to that which is unknown, utilizing all other evidence available for the study area: topo and thematic maps, photographs or data taken at ground level, or a visit to the site itself.
Since remote sensing is generally less expensive than field surveys, especially as labour costs increase and computer processing becomes more accurate, efficient and less expensive, the trend to replace "contact" methods of assessment with remote sensing will continue. By correlating remote sensing with field data, measurements such as vegetation densities or groundwater levels can be extrapolated to similar areas using remote sensing alone. In the same manner, remote sensing can be used to monitor changes in sites over time once a connection has been made between image patterns and what they represent on the ground. Changes can be detected from imagery, but the cause of the change must be inferred by the user.
Classification is either "supervised" or "unsupervised". In supervised classification, "training sets" of both remotely sensed and ground truth data are used to categorize patterns on imagery. When the same patterns occur elsewhere on the image, they are judged to belong to the same category. Unsupervised classification is used when ground truth data are not available. Instead, natural groupings of pattern or light on an image are identified and assigned to categories based on imagery characteristics alone. Classification can be done manually using photographs, or by applying statistical techniques such as cluster analysis and principal component analysis to digital satellite data. The eye is good at interpretation but brightness and texture are more accurately identified with the assistance of a computer. Classification accuracy is increased by decreasing the spatial resolution of remotely sensed data (making it "fuzzier" and thus more uniform), decreasing the number of classes, or increasing the information available to the classifier.
Geographic information systems (GIS) provide a relatively new tool for classification. The concept is the same as flipping backwards and forwards between several maps and photos in the hopes that one's brain will combine the information. Unfortunately, when more than two or three maps are visually compared, the mind tends to overload Computer GISs can superimpose spatial data from thematic maps, digital terrain models, and imagery to yield new maps showing, for example, areas with high erosion potential (from combining climate, topography and soils information). An additional benefit of the digital terrain data is that "maps" developed from this data can be given a 3-dimensional look by changing the viewing angle and adding shadows.
To collect ground truth data for classification purposes is a
"question of how much ground checking will be required to
produce a result comparable in objective accuracy to a ground
survey" (Barrett and Curtis 1982:120). Sampling schemes
should be based on the statistical sampling designs given in Chapter
1. A stratified sampling design is preferred, with samples taken
within each classification. Curran (1985) recommends at least
50 sample points per class, which is impractical if the site access
is difficult or the object of study changes with time. Limitations
of "representative" sampling, discussed in section 1.2.4
also apply to the selection of sampling site. Variations range
from carefully locating a representative, accessible site on an
image which is then studied in the field - to "observation
while moving at a rapid pace on adjacent roads" (Myers, 1975
in Curran, 1985). These methods are relatively fast and utilize
the field knowledge of the investigator; however, its drawback,
again, is that the data are biased and statistically invalid.
When classification is a subjective process to begin with, there
is probably sufficient "truth" provided by judgemental
sampling.
 Go back to the Horizons page
|
 Go forward to Photogrammetry
|