Understanding inference–evidence relationships in recognition tasks is fundamental science, and experimenting with these relationships using ‘real world’ problems makes them more tangible. It is the richness of geographical knowledge on one hand, and its visual representability on the other (e.g. Golledge, 2002), which may facilitate such experimentation. Recognition of geographical phenomena in wide areas from remote sensing images is a most complex problem due to its high dimensionality (spectral, temporal, spatial). The objectivity, consistency and accessibility of remote sensing data make it a good case study for assessing different recognition concepts within the framework of scientific reasoning. Addressing the problem of reasoning, remote sensing recognition raises issues of types of evidence, methods of inference, ways of representing and implementing context and how to create a semantic linkage between the object of recognition and the recognition process itself. These concepts, issues, questions and problems have received major attention in recent years within the fields of knowledge discovery and in addressing the ontology of geographical information (Frank, 2001; Winter, 2001; Visser et al., 2002; Miller and Han, 2001; Gahegan, 2001; Wachowicz, 2001). We have aimed at contributing to the understanding of these issues by assessing relationships between evidence and inference utilizing the case study of agricultural land-use mapping.
Initially, ways of creating a semantic linkage between objects and processes of recognition had to be addressed. It was suggested that the terms “unresolved complexity”, “required recognition energy” and “evidential effort” may share a common representation The interoperability between these terms is most significant, since “unresolved complexity” represents the level of the recognition challenge, whereas “evidential effort” represents the work needed to form the knowledge basis, which will facilitate recognition. It is therefore possible to suggest that the distance between them represents the sophistication or intelligence required for resolving this complexity. By forming a common representation it was possible to quantitatively estimate each of these components, and furthermore to quantify the contribution of the inference toward the resolution of the recognition problem.
Inference and evidence are two central elements in the reasoning process. Assessment of relationships between them is extended when considering implicit versus explicit evidence and domain-dependent versus independent inference. Domain-independent inference (DII) represents one of the common intelligence capabilities founded on general principles of induction, deduction or abduction. The highest level of intelligence is gained in resolving most complex problems from a most redundant data/information of the implicit type. Simulating different weights enabled the identification of relationships between the parameters defining the workloads and consequently quantifying the evidential effort. The role of the repetitive use of statements versus the ‘cost’ of sequential procedures was demonstrated. The bottom line suggests that DII may have significant gain when the production of soft evidence requires a third or less effort than that required for producing hard explicit evidence. Such relationships must be further analysed conceptually and experimented before general conclusions regarding the applicability of the approach to a wide range of phenomena types and complexities can be reached. However, it must be emphasized that the gain identified here for DII in general terms must be attributed specifically to the Dempster-Shafer theory of evidence. In our opinion it facilitates an important shift from a search for the principle attributes/components/ordinates to approaches integrating vast amounts of implicit data by adjusting the context and by being resilient to contradicting and heterogeneous evidence. It has been shown in our case that utilizing as many as 1278 conditional statements improved the recognition accuracy and reliability in test areas not specifically represented in the training areas. Exploitation of important knowledge and maximization of their utilization in identifying important patterns and relationships depend to a large extent on the availability of intelligent inference methods capable of handling data containing errors, contradictions, uncertainties and gaps.
Knowledge Based Systems 
