stored individually, the files become very large. A simple way to reduce the
required storage (and one of the oldest) is to process each row of the raster
data set from left to right, recording only when the attribute value changes
and the number of cells following the change to the right. For example, if a
row is 100 cells long and cell 1-20 has the value 156, cells 21-78 have the
attribute value 123, and cells 79-100 have the attribute value 156 again, the
run-length encoded (RLE) raster storage would only store 156:20; 123:59;
and 156:21. Other systems are more complicated, but even more efficient.
One of the most interesting storage formats is the quad-tree format which
works like the RLE approach, but puts areas into a hierarchy of cell value.
For example, an agricultural raster data set representing types of crops could
distinguish crops at the highest level by the genus, at the next level down in
the quad-tree hierarchy it could show the Linnean classification family, and
at the third level of the quad-tree it could show individual species. The quad-
tree is very efficient and very fast, but changes to the hierarchy can be very
complicated and require a great amount of processing.
Network-Based Geographic Representation
The network geographic representation type is usually considered to be a
subtype of the position-based geographic information type, but is distinct
because of its special properties for representing topological relationships.
The network geographic representation type uses nodes and links,
which correspond to nodes and chains in the vector position-based geo-
graphic representation type. The distinction is that nodes in the network
store information about possible connections (e.g., possible turns at an inter-
section) and links store the information about how nodes are topologically
connected (e.g., Chicago is connected to St. Louis by Interstate 55). Topolog-
ical information is extremely helpful for vector-based network GI.
Nodes can be added with coordinates from a coordinate system and
with additional points with coordinates to define the shape of the networks,
for example, situating Chicago and St. Lous on the map in a geographically
correct arrangement. However, many networks are represented without this
location information, allowing the map to be very simple and easily read
(e.g., public transportation maps). (See Plate 6, the London Underground
Field-Based Geographic Representation
For the representation of nondiscrete, mainly environmental, properties
including soil moisture, soil pH, or the distribution of airborne particles and
substances including ozone, dust, or pollen, fields are the ideal GI represen-
Conceptually, fields are nondiscrete, meaning no precise and accurate
boundaries can be made between soil pH 6.7 and 6.8, and the properties of a
field can be modeled using geostatistical techniques that take these relation-