Text Mining Analysis of Land Cover Semantic Overlap

Abstract

This chapter explores the origins and impacts of semantic variation using the example of land cover. It examines the origins of semantic variation in land cover mapping and the philosophical process of categorization, comparing what might be called top-down and bottom-up nomenclatures. In doing so, it illustrates that land cover classifications, as with many geographical concepts, are vague, imprecise, and socially constructed: they represent a coming together of a particular world view (weltanschauung). The context of ubiquitous representation in digital cartographic products and the origins of semantic variation in land cover mapping are examined. This chapter examines the origins of semantic variation in land cover mapping. The chapter considers the influences of specific factors on geographic representation and the need to abstract the infinite complexity of the real world into spatial databases. In so doing, it describes how the need for generalization processes such as abstraction and aggregation are a series of choices. Choices are shown to relate to the commissioning and policy background (who paid for it?), observer variation (what did you see?), institutional variation (why you see it that way?), and variation in measurement variation (how was it recorded?). This problem is tackled using a raw text mining approach that seeks to characterize the semantic overlap between semantically discordant datasets and then to integrate in a land cover change. The results of the text mining are compared with human experts and shown to be more efficient at characterizing the consistency between two land cover maps in the context of land cover change and semantic discordance. A number of critical research areas are identified relating to dynamic metadata and formal ontologies, linguistic issues, and different semantic latency approaches.