114
Matt Duckham and Mike Worboys
To our knowledge, the study by Duckham and Worboys [22] is the only research
in the geographical domain that adopts an extensional approach to automating information fusion (although it is not the only work to acknowledge the importance
of extensions in the representation of geographical knowledge, e.g. [8, 9]). Geographical information is a richly structured and voluminous source of instances upon
which to base inductive reasoning processes, more so than many other types of information source. In this respect, it is well suited to extensional approaches to automated information fusion. However, the problems of unreliable inference introduced
in Sect. 6.2.2 are exacerbated in the geographical domain because uncertainty is an
endemic feature of geographical information. Applying an unreliable reasoning process to uncertain data could potentially generate information that is degraded to the
point of being meaningless. Consequently, following a closer look at using induction as a basis for automated geographical information fusion in Sect. 6.3, we turn
to the issues of unreliability in the reasoning process (Sect. 6.4) and reasoning under
uncertainty (Sect. 6.5).
6.3 ROSETTA: Automated Extensional Geographical
Information Fusion
At the core of an extensional approach to automated geographical information fusion is the process of inferring semantic relationships from spatial relationships. As
already discussed, this process is an example of inductive inference: reasoning from
specific cases to general rules. As an analogy, archaeologists were able to decipher
the meaning of ancient Egyptian hieroglyphs following the discovery of the Rosetta
Stone, a 2nd-century tablet that contained the same official decree in both Egyptian
(hieroglyphs and text) and Greek (text). Before the discovery and subsequent analysis of the Rosetta Stone, all attempts to decipher hieroglyphs were unsuccessful
and Egyptian hieroglyphics were considered to be merely primitive picture writing.
Only by comparing examples (extensions) of the Greek text with Egyptian text and
hieroglyphs on the Rosetta Stone were archaeologists able to correctly infer a “dictionary” (intensions) for translation between these different information sources. In a
similar way, the extensional approach to geographical information fusion constructs
a shared “dictionary” for translating between the ontologies of the different information sources, based on the relationship between the spatial extents of the categories
used in those information sources. In the remainder of this chapter, we use the term
“rosetta” to refer to the extensional approach to automating geographical information fusion.
To illustrate, Fig. 6.1 contains a much simplified example of a rosetta-based
fusion. In Fig. 6.1, each data set comprises an extensional component (the mapped
spatial data) and an intensional component (the ontology for that spatial data). On
the left-hand side of Fig. 6.1, the intension for data set A contains the categories
Forest and Built-up area, while the extension contains two regions, one of each category. Similarly, on the right-hand side of Fig. 6.1, data set B contains the intensions
Woodland and Urban along with a map of the spatial extensions of the Woodland
Matt Duckham and Mike Worboys
To our knowledge, the study by Duckham and Worboys [22] is the only research
in the geographical domain that adopts an extensional approach to automating information fusion (although it is not the only work to acknowledge the importance
of extensions in the representation of geographical knowledge, e.g. [8, 9]). Geographical information is a richly structured and voluminous source of instances upon
which to base inductive reasoning processes, more so than many other types of information source. In this respect, it is well suited to extensional approaches to automated information fusion. However, the problems of unreliable inference introduced
in Sect. 6.2.2 are exacerbated in the geographical domain because uncertainty is an
endemic feature of geographical information. Applying an unreliable reasoning process to uncertain data could potentially generate information that is degraded to the
point of being meaningless. Consequently, following a closer look at using induction as a basis for automated geographical information fusion in Sect. 6.3, we turn
to the issues of unreliability in the reasoning process (Sect. 6.4) and reasoning under
uncertainty (Sect. 6.5).
6.3 ROSETTA: Automated Extensional Geographical
Information Fusion
At the core of an extensional approach to automated geographical information fusion is the process of inferring semantic relationships from spatial relationships. As
already discussed, this process is an example of inductive inference: reasoning from
specific cases to general rules. As an analogy, archaeologists were able to decipher
the meaning of ancient Egyptian hieroglyphs following the discovery of the Rosetta
Stone, a 2nd-century tablet that contained the same official decree in both Egyptian
(hieroglyphs and text) and Greek (text). Before the discovery and subsequent analysis of the Rosetta Stone, all attempts to decipher hieroglyphs were unsuccessful
and Egyptian hieroglyphics were considered to be merely primitive picture writing.
Only by comparing examples (extensions) of the Greek text with Egyptian text and
hieroglyphs on the Rosetta Stone were archaeologists able to correctly infer a “dictionary” (intensions) for translation between these different information sources. In a
similar way, the extensional approach to geographical information fusion constructs
a shared “dictionary” for translating between the ontologies of the different information sources, based on the relationship between the spatial extents of the categories
used in those information sources. In the remainder of this chapter, we use the term
“rosetta” to refer to the extensional approach to automating geographical information fusion.
To illustrate, Fig. 6.1 contains a much simplified example of a rosetta-based
fusion. In Fig. 6.1, each data set comprises an extensional component (the mapped
spatial data) and an intensional component (the ontology for that spatial data). On
the left-hand side of Fig. 6.1, the intension for data set A contains the categories
Forest and Built-up area, while the extension contains two regions, one of each category. Similarly, on the right-hand side of Fig. 6.1, data set B contains the intensions
Woodland and Urban along with a map of the spatial extensions of the Woodland