4 Integration of Remotely Sensed Data into Geographical Information Systems
71
can be easily read and imported by most GIS. The second step is to transform
imported lattice data into a polygon vector layer. This is achieved using the SVF
format. Built-in routines are available in a vector GIS to convert lattice format into
SVF format (e.g., LATTICEGRID routine of Arc/Info), and then SVF format into
a polygon vector layer (e.g., GRIDPOL Y) (ESRI, 1991). Polygons in the resulting
vector layer will be built from groups of contiguous pixels having the same pixel
value. Normally, these routines create lines along pixel borders thus forming
closed polygons. Mattikalli (1995) employed this approach to integrate satellite
data derived from both, fine and coarse resolution sensors with data digitized from
maps.
This methodology has both, advantages and minor limitations. Main advantages
are that it does not require sophisticated computer hardware and software systems,
and that the result does not introduce new errors because no data interpolation is
involved in the conversion. One limitation is that the process is relatively data
expensive (in terms of storage) since polygon boundaries follow pixel borders,
which could be a major difficulty for use with raw satellite images. However, this
is not a severe problem for watershed applications that normally use classified
image products.
Errors contained in the original data are carried-on and propagate through geographical processing performed in a hydrological analysis. Although it is not possible to eliminate errors in an integrated application, they can at least be managed
and kept to an acceptable minimum. The following section briefly discusses the
common and most obvious sources of errors and their propagation.
4.2.3
Errors Associated with Geographical Processing
Spatial data accuracy is governed by user requirements, inherent characteristics of
data source, and instruments used to create digital data. Errors can arise at every
stage of using a GIS, from collection of the original data to the output and use of
resulting information (Walsh et aI., 1987; Lunetta et aI., 1991). Burrough (1990)
discusses various sources of errors in a GIS. Needless to say, positional accuracy
in a GIS cannot exceed those of the original data source whether it is a large-scale
map sheet that has been digitized or data collected from an orbital remote sensor.
Original maps and their cartographic features inevitably contain errors, no matter
how small, and precision of the input data depends on the scale of the original
map. GIS layers derived from maps of varying scales are subject to such errors.
Remotely sensed data integrated into a GIS are no exception. Remotely sensed
data are invariably classified to derive thematic products. Although, classifications
often result in a fair degree of accuracy, such products will have some errors associated with them. In addition, remotely sensed data will be associated with errors
due to geometric correction and rectification. Positional accuracy of spatial entities recorded in a GIS are of critical importance because of the problems that can
be generated during overlay operations. Within the limits of accuracy of a given
input layer, a variety of accuracy levels may be utilized depending upon the application needs. Information derived (e.g., area calculations rather than precise de-
71
can be easily read and imported by most GIS. The second step is to transform
imported lattice data into a polygon vector layer. This is achieved using the SVF
format. Built-in routines are available in a vector GIS to convert lattice format into
SVF format (e.g., LATTICEGRID routine of Arc/Info), and then SVF format into
a polygon vector layer (e.g., GRIDPOL Y) (ESRI, 1991). Polygons in the resulting
vector layer will be built from groups of contiguous pixels having the same pixel
value. Normally, these routines create lines along pixel borders thus forming
closed polygons. Mattikalli (1995) employed this approach to integrate satellite
data derived from both, fine and coarse resolution sensors with data digitized from
maps.
This methodology has both, advantages and minor limitations. Main advantages
are that it does not require sophisticated computer hardware and software systems,
and that the result does not introduce new errors because no data interpolation is
involved in the conversion. One limitation is that the process is relatively data
expensive (in terms of storage) since polygon boundaries follow pixel borders,
which could be a major difficulty for use with raw satellite images. However, this
is not a severe problem for watershed applications that normally use classified
image products.
Errors contained in the original data are carried-on and propagate through geographical processing performed in a hydrological analysis. Although it is not possible to eliminate errors in an integrated application, they can at least be managed
and kept to an acceptable minimum. The following section briefly discusses the
common and most obvious sources of errors and their propagation.
4.2.3
Errors Associated with Geographical Processing
Spatial data accuracy is governed by user requirements, inherent characteristics of
data source, and instruments used to create digital data. Errors can arise at every
stage of using a GIS, from collection of the original data to the output and use of
resulting information (Walsh et aI., 1987; Lunetta et aI., 1991). Burrough (1990)
discusses various sources of errors in a GIS. Needless to say, positional accuracy
in a GIS cannot exceed those of the original data source whether it is a large-scale
map sheet that has been digitized or data collected from an orbital remote sensor.
Original maps and their cartographic features inevitably contain errors, no matter
how small, and precision of the input data depends on the scale of the original
map. GIS layers derived from maps of varying scales are subject to such errors.
Remotely sensed data integrated into a GIS are no exception. Remotely sensed
data are invariably classified to derive thematic products. Although, classifications
often result in a fair degree of accuracy, such products will have some errors associated with them. In addition, remotely sensed data will be associated with errors
due to geometric correction and rectification. Positional accuracy of spatial entities recorded in a GIS are of critical importance because of the problems that can
be generated during overlay operations. Within the limits of accuracy of a given
input layer, a variety of accuracy levels may be utilized depending upon the application needs. Information derived (e.g., area calculations rather than precise de-