cells of rasters can contain either continuous or discrete values. As mentioned earlier, image data is continuous. However, when we classify an image into information, the resulting values can be either continuous, as in an elevation model (figure 2.5), or discrete, as in a land use map such as that shown in figure 2.6. Unlike continuous data, discrete data classes cannot be mathematically divided more finely. Discrete information can take on only finite predefined values such as “tank,” “lake,” “urban,” “forest,” “building,” or “agriculture.” Rasters of discrete values represent information that has been classified from image data; they are no longer considered images, but are rather now raster format maps.
Figure 2.5. An example of a continuous raster in the form of a digital elevation model (DEM)
Figure 2.6. A thematic map showing discrete land-cover classes present in the Coastal Watershed in southeastern New Hampshire created from Landsat 8 imagery
Bands
Imagery measurements are collected and stored in raster bands. For example, a panchromatic image raster includes only a single band of measurements, shown as a single layer in figure 2.3, and is typically shown in grayscale. Multispectral imagery contains several bands of measurements, as shown in figure 2.7. Figure 2.8 shows a portion of Landsat imagery over Sonoma County, California. The numerical values of the cells of three bands of the seven-band image are displayed. When the red, green, and blue bands are displayed in the red, green, and blue colors of a computer screen they create the natural-color image of figure 2.8.
Figure 2.7. Multispectral data. If more than one type of measurement is collected for each cell, the data is called multispectral, and each type of measurement is represented by a separate band.
Figure 2.8. The numerical values of three bands of Landsat imagery over a portion of Sonoma County, California.
Hyperspectral data contains 50 to more than 200 bands of measurements and is usually represented as a cube of spectral values over space (figure 2.9). Image cubes are also used to bring the temporal dimension into a set of images, as when multiple Landsat images are analyzed of the same area over time.
Figure 2.9. A hyperspectral data cube captured over NASA’s Ames Research Center in California. Hyperspectral data includes 50 to more than 200 bands of measurements. Source: NASA
Figure 2.10. The impact of raster cell size on the level of detail depicted. The larger the cell, the less discernible detail. In this example a car is represented by three different image cell sizes but displayed at the same scale. The smaller the cell, the more information available to identify the rectangle of eight large reddish pixels on the right as a red sedan on the left.
Cell Size
The cell size, or spatial resolution, of a raster will determine the level of spatial detail displayed by the raster. Figure 2.10 illustrates the effect of cell size on spatial resolution. The cell must be small enough to capture the required detail but large enough for computer storage and analysis to be performed efficiently. More features, smaller features, or greater detail in the extent of features can be represented by a raster with a smaller cell size. However, more is not always better. Smaller cell sizes result in larger raster datasets to represent an entire surface; therefore, there is a need for greater storage space, which often results in longer processing time.
Choosing an appropriate cell size is not always simple. You must balance your application’s need for spatial resolution with practical requirements for quick display, processing time, and storage. Essentially, in a GIS, your results will only be as accurate as your least accurate dataset. The more homogeneous an area is for critical variables, such as topography and land use, the larger the cell size can be without affecting accuracy.
Determining an adequate cell size is just as important in your GIS application planning stages as determining what datasets to obtain. A raster dataset can always be resampled to have a larger cell size; however, you will not obtain any greater detail by resampling your raster to have a smaller cell size. Chapter 3 discusses cell size and image spatial resolution in more detail.
How Is Imagery Used in a GIS?
The three primary uses of imagery in a GIS are
1 1.as a base image to aid the visualization of map information, as shown in figure 2.11
2 2.as an attribute of a feature. For example, an image of vegetation taken from the ground may serve as an attribute of a vegetation survey point displayed on a map, as shown in figure 2.12
3 3.as a data source from which information is extracted through the process of image classification. For example, imagery may be interpreted by image analysis to determine the current state of situations for disaster response, environmental monitoring, or military planning. Imagery can also be transformed into informational map classes through manual interpretation or semi-automated classification.
The focus of much of this book is on the third use—image classification, which is the process of utilizing imagery in a GIS to produce maps.
Figure 2.11. Imagery as a base image. This figure shows airborne infrared imagery as a base image with parcel boundaries (in yellow) and field data points (in green). (esriurl.com/IG211). Source: Sonoma County Agriculture Preservation and Open Space District
Figure 2.12. A field-captured image as an attribute of the survey point geodatabase. Source: Sonoma County Agriculture Preservation and Open Space District
Image Classification — Turning Data into Map Information
To simplify and make sense
