GEOG 486
Cartography and Visualization

Raster Data


We introduced the basics of the raster data model in Lesson 1, Part VI: Objects vs. Fields (and Vector vs. Raster). In this part, we will look at raster data in more detail and think about how different structures may have an impact upon your work as a cartographer.

Just to refresh your memory, the raster data model describes the world as a field over which the spatial distribution of some phenomenon varies. The geographical primitive that we use with this data model is a pixel. Pixels are commonly rectangular or square in shape and can be nested within each other to represent the phenomenon of interest at different levels of detail (i.e., resolutions). The raster data model is also different from the vector data model in that only one corner of the dataset needs to have an absolute georeference (i.e., coordinate in latitude and longitude or some other coordinate system); the location of all other pixels can be calculated on an as needed basis from their relative position from the georeferenced point and the resolution of the pixels.

Raster data are commonly used for three main types of mapping applications: mapping distributions of natural phenomena such as vegetation type (e.g., from a classified satellite image), as a backdrop for reference maps (e.g., digital orthophoto quadrangles created from aerial photography), or they may also be used to generate other types of surface representations that are displayed in maps (e.g., shaded relief).

The main factor that will have an impact upon your work as a cartographer with raster data is the resolution of the datasets with which you are working. As we briefly described in Lesson 1, Part VI: Objects vs. Fields (and Vector vs. Raster), raster data file sizes increase with increasing resolution. Typically, an increase of 2 times the resolution (e.g., going from 60m pixels to 30m pixels; see Figure, below) will increase the file size by a factor of four.

A diagram to show a comparison between datasets with different numbers of pixels, which effects disk space.
Figure In a simple raster file structure, each pixel takes up a fixed amount of disk space. Increasing the resolution of a dataset increases the number of pixels, and thereby the amount of disk space needed to store the file. In the example above, the dataset at the right will result in a file that is four times larger than the dataset at the left.

Thanks to today's relatively fast desktop computers, file size is much less of an issue than it was previously; however, very large file sizes of high-resolution imagery (with pixels representing areas of less than 1m2 on the ground) can still have a substantial impact on your computer's performance. There are several different types of raster data structures that are designed to reduce file sizes (e.g., run-length encoding and quadtrees; see Figure, below), these compression methods are most useful for raster data that do not contain a large amount of pixel to pixel variation (as most air-photos or satellite images do).


A graphic illustration of a simple raster format.visual spaceA graphic illustration of a raster format with its corresponding run-length encoding.visual spaceA graphic illustration of a raster with quadtree formatting.
Figure In a simple raster format, as in the top image, the computer stores one piece of information per pixel (256 pieces in this example). Run-length encoding can reduce the number of pieces of information needed if there are strings of pixels with identical values. This method stores the pixel value and then the pixel position where the "run" ends. In the middle image (with its corresponding run-length encoding at the right), 188 pieces of information are needed to store the example data. Quadtrees (bottom image) reduce the amount of data storage needed to store a particular dataset by recursively dividing the space into quarters until all pixels in a given quarter have the same value. In this example, the quadtree format requires 160 pieces of information. By looking at the above images, you should be able to imagine that in continuously varying raster surfaces, neither run-length encoding nor quadtrees would be particularly good at reducing the size of the raster data file.

If you are using raster data as a backdrop image for your map, the resolution will also have an impact upon the visual appearance of objects within the image. At lower resolutions, many objects may take on a 'stair-step' appearance that can be distracting for the map reader, and at worst, can even prevent the map reader from being able to identify the feature (see Figure, below).


A section of satellite imagery with a one meter resolution, which produces an acceptable image.visual spaceA section of satellite imagery with a five meter resolution, which produces a blocky, less acceptable image.
Figure Ikonos satellite imagery ( Space Imaging, Inc.) with a 1-meter resolution (top) could easily be used as a backdrop for mapping at this particular scale. Imagery of the same area with a 5-meter resolution, however, results in blocky features that are difficult to identify and may detract from a map reader's interpretation of other features if they are combined with the image in a map.

The resolution of raster terrain data can also affect the appearance of shaded relief: if the source data are not at an appropriate scale for the final map product, the shaded relief can look blocky (see Figure, below).

A shaded relief employing data with a 60 meter resolution, which results in a blocky image.
Figure Here, we have created a shaded relief image at the same map scale as the shaded relief images that are shown in Shaded Relief. However, in this example, we have used data with a 60m resolution rather than the 30m resolution used in the earlier examples. Notice the blocky appearance of the shaded relief here; this coarser-resolution data would be more appropriately used for creating shaded relief for smaller-scale maps.

When you need to reproject raster data into a different projection, it is also important to be aware of the potential impacts of different resampling methods on the data. For example, if you have a thematic raster data set (e.g., land use or soil type), you should choose a nearest neighbor interpolation method rather than a bilinear or cubic convolution method (see Figure, below). In cases where you need to use an equal area projection for the map you are creating, also be aware that reprojecting raster data can actually introduce errors into the data set, with this problem being most pronounced at lower resolutions (Usery et al. 2003).

A resampling map made with a nearest neighbor algorithm.visual spaceA resampling map made with a bilinear interpolation.
Figure In the map at the left, we used a nearest neighbor algorithm for resampling that occurred when the data were reprojected, while we used a bilinear interpolation for the map at the right. Notice the introduction of resampling-related artifacts into the dataset in the map on the right (visible in the narrow band of multi-colored pixels). Here, the bilinear resampling is altering the values of these edge pixels to create values that were not present in the original dataset, while the nearest neighbor resampling algorithm uses only values that were present in the original data.

Recommended Readings

If you are interested in investigating this subject further, I recommend the following:

  • Usery, E. L., Finn, M. P., Cox, J.D., Beard, T., Ruhl, S. and M. Bearden. 2003. "Projecting global datasets to achieve equal areas." Cartography and Geographic Information Science. 30(1).