GIS (geographic information system)

Using different regions of the electromagnetic spectrum to identify and to do analysis of remotely sensed features.

            Remote sensing sensors (Landsat, SPOT, AVIRIS, AVHRR, LIDAR, SAR, etc.) record the relative brightness of an area over specific portions of the electromagnetic spectrum.  All sensors have spectral sensitivity limitations; this is referred to as spectral resolution.  No single sensor is sensitive to all wavelengths of the electromagnetic spectrum.   Recorded wavelengths are referred to as bands.  The number of bands varies depending on the sensor system (multispectral, hyperspectral, radar).   Displaying of a remote sensing image on a computer monitor is limited to 3 bands.  Selected bands are shown consecutively through the three color monitor guns (red, green, and blue).  This produces a false image.  Band color combinations are dependent upon the type of feature analysis being performed.

Reflectance values are a result of "...energy reflected and emitted back from an object that is detected by a sensor. The measure of reflected energy is referred to a radiometric resolution.  By analyzing energy received by the sensor, information about features can be derived." (Arnoff, p. 63). The energy that is reflected or emitted back represents characteristic of a feature at that particular moment. All features have unique reflectance characteristics. This is useful when identifying features represented within any type of image (pan chromatic, remotely sensed, or radar). Reflectance values can be easily imported into a GIS.

At one time it was thought that each object had its own spectral signature.  This would mean that a birch tree would have one reflectance value and a maple tree would have a totally separate reflectance value. In the 1970's it was realized that this could not be achieved for two main reasons: 1.there are a variety of factors that may change an objects reflectance patterns such seasonal changes, environmental moisture content, and 2. data format.  When dealing with raster based information mixed pixels (mixels) are inevitable.  All sensors have in inherent limitation to just how small of an object on the Earth s surface can be identified from its surroundings.  The measure of size is referred to as spatial resolution.  Spatial resolution reflects the smallest object that can be detected by a sensor.  As an example, Landsat TM has a spatial resolution of 30 X 30 meters.  The sum of all of the spectral reflectance of all features within the 30 X 30 meter footprint comprises a spectral response pattern that is detected by the sensor.  If an operator want to identify features that are less then 30 X 30 meters, a different sensor with a resolution <30 meters must be selected. 

This is a sophisticated program that uses statistical techniques to discriminate between land cover types from remotely sensed imagery (i.e. determining if an area is a forest or wetland using reflectance values). 

Land cover classification from remotely sensed imagery that requires minimal operator input is referred to as unsupervised classification. 

Land cover classification from remotely sensed imagery that requires significant operator input is referred to as supervised classification. 

This is a program procedure that is based on probability analysis to identify or classify what features are. It is an image classification algorithm.  When used it says there is a 60% (or whatever the percentage is) that the object is what it is. This value can be calculated by performing statistical measures and weighting the data. Today, probability analysis along with image classification algorithms are used to distinguish features and analyze data compared to when object recognition was the preferred method. This new method became known as multispectral pattern recognition; the different spectral responses are used to tell about the image versus the shape of features.