Error Modeling and Management for Data
357
themselves in a variety of error forms (e.g., geometric, attribute and so on) which has
been extensively studied by many researchers (Hunter and Beard, 1992; Goodchild et
al., 1992; Chrisman, 1991). One of the most basic forms of error classification
distinguishes between “geometric” error, or error in positional features such as points
and lines, and “attribute” error, or errors in the values of thematic attributes. Chrisman
(1989) refers to these two forms of errors as “positional” and “attribute” errors
respectively while Bedard (1987) calls them “locational” and “descriptive” errors.
However, it should be remembered that spatial data (such as coordinates of points) and
aspatial data (such as landuse cover) are both attributes of a spatial entity.
ERROR MODELING
The second stage of the strategy is concerned with methods of assessing accuracy levels
and error modeling. For aspatial error, the measurement of error is frequently done by
building confusion matrices. This method is common when evaluating the performance
of a classification algorithm for digital images. While confusion matrices are
appropriate for categorical data, a variety of methods have been proposed for assessing
spatial error (Alesheikh, 1997). In addition, different applications will place different
priorities on the various uncertainty forms. While positional errors are the focus of this
chapter and more generally the focus of spatial modelers, attribute errors are a major
concern in other applications and probably the combination of the two may be
significant. In addition, other forms of uncertainty such as logical consistency may
need more attention in applications where topological information is needed.
This step is also concerned with evaluating and modeling the propagation of
error as a consequence of applying GIS operations that involve spatial data. The
magnitude of errors may be drastically changed due to different operations that the
data are subjected to. Examples of error propagation may be found when determining
the uncertainty of a straight line object in a GIS and solving the point-in-polygon
problem. These issues will be elaborated upon in the following sections.
ERROR COMMUNICATIONS
The concept of reporting the quality of a product is not new, with some of the earliest
forms being positional accuracy statements as found on hardcopy topographic maps.
To realize the fitness of data for a specific application the accuracy of the data (i.e.
metadata) should be provided. It should be the producer’s responsibility to provide the
necessary information to allow users to determine whether the product is suitable for
their use in the first place - which brings up the notion of “truth in labeling”.
Though many data transfer standards have been in place for reporting errors,
the US Spatial Data Transfer Standard appears to be the most complete (Fegeas et al.,
1992). Regarding the data accuracy and quality, the standard has a specific provision
for information about the following components: data lineage, positional accuracy,
attribute / temporal accuracy, logical consistency; and completeness. The rationale is
357
themselves in a variety of error forms (e.g., geometric, attribute and so on) which has
been extensively studied by many researchers (Hunter and Beard, 1992; Goodchild et
al., 1992; Chrisman, 1991). One of the most basic forms of error classification
distinguishes between “geometric” error, or error in positional features such as points
and lines, and “attribute” error, or errors in the values of thematic attributes. Chrisman
(1989) refers to these two forms of errors as “positional” and “attribute” errors
respectively while Bedard (1987) calls them “locational” and “descriptive” errors.
However, it should be remembered that spatial data (such as coordinates of points) and
aspatial data (such as landuse cover) are both attributes of a spatial entity.
ERROR MODELING
The second stage of the strategy is concerned with methods of assessing accuracy levels
and error modeling. For aspatial error, the measurement of error is frequently done by
building confusion matrices. This method is common when evaluating the performance
of a classification algorithm for digital images. While confusion matrices are
appropriate for categorical data, a variety of methods have been proposed for assessing
spatial error (Alesheikh, 1997). In addition, different applications will place different
priorities on the various uncertainty forms. While positional errors are the focus of this
chapter and more generally the focus of spatial modelers, attribute errors are a major
concern in other applications and probably the combination of the two may be
significant. In addition, other forms of uncertainty such as logical consistency may
need more attention in applications where topological information is needed.
This step is also concerned with evaluating and modeling the propagation of
error as a consequence of applying GIS operations that involve spatial data. The
magnitude of errors may be drastically changed due to different operations that the
data are subjected to. Examples of error propagation may be found when determining
the uncertainty of a straight line object in a GIS and solving the point-in-polygon
problem. These issues will be elaborated upon in the following sections.
ERROR COMMUNICATIONS
The concept of reporting the quality of a product is not new, with some of the earliest
forms being positional accuracy statements as found on hardcopy topographic maps.
To realize the fitness of data for a specific application the accuracy of the data (i.e.
metadata) should be provided. It should be the producer’s responsibility to provide the
necessary information to allow users to determine whether the product is suitable for
their use in the first place - which brings up the notion of “truth in labeling”.
Though many data transfer standards have been in place for reporting errors,
the US Spatial Data Transfer Standard appears to be the most complete (Fegeas et al.,
1992). Regarding the data accuracy and quality, the standard has a specific provision
for information about the following components: data lineage, positional accuracy,
attribute / temporal accuracy, logical consistency; and completeness. The rationale is