Introduction
In Geographic Information Systems (GIS), a persistent challenge exists in bridging the gap between raw spatial data—represented as pixels or coordinates—and the high-level natural language humans use to describe the world. Terms like "river," "lake," and "pond" are inherently vague. When does a widening stream become a lake? When does a narrow channel become a canal? This paper introduces a formal framework designed to classify inland water features by accounting for the "Sorites vagueness" of human language, utilizing a combination of Formal Concept Analysis (FCA) and Supervaluation Semantics.
The Challenge of Vagueness in GIS
Traditional logical systems are often too rigid for geographic classification. A single-meter change in width shouldn't abruptly flip a classification from "stream" to "river," yet standard boolean logic requires such sharp boundaries. While fuzzy logic is a common alternative, it often struggles with maintaining reliable logical inference.
The authors argue that vagueness is dual-layered: it affects both classification (what a feature is) and individuation (where one feature ends and another begins). To address this, they propose a system that respects different "standpoints" or interpretations of thresholds, allowing for a more flexible yet mathematically rigorous ontology.
Methodology: FCA and Supervaluation
The framework relies on two primary theoretical pillars to structure geographic knowledge:
- Formal Concept Analysis (FCA): This method is used to construct the "skeleton" of the ontology. By taking a set of water feature types and their defining attributes (such as linearity, flow, and origin), FCA generates a complete lattice. This lattice visually and mathematically maps the hierarchical relationships and overlaps between concepts like "River," "Canal," and "Reservoir."
- Supervaluation Semantics: To handle the inherent vagueness of these concepts, the authors employ "precisifications." A statement about a vague geographic feature is considered "super-true" only if it remains true under every possible precise interpretation (or standpoint) of the underlying thresholds, such as the exact width required to define a "large" water body.
Practical Implementation: Bridging Geometry and Semantics
To move from theory to application, the authors utilize Medial Axis Transformation (Skeletonization). This image processing technique reduces 2D water features into 1D "skeletons" that preserve the distance to the nearest shore at every point. This allows the system to dynamically calculate width and linearity, which are then fed into a Prolog-based logic engine.
The resulting system can take a raw bitmap of a hydrographic network and automatically segment it into qualitatively distinct features. For example, it can identify which parts of a complex network are "river stretches" versus "ponds" or "lake-like" bulges, even when they are physically connected.
Conclusion and Key Insights
The integration of FCA and Supervaluation Semantics provides a robust way to handle subjective geographic definitions without sacrificing logical consistency. By treating thresholds as parameters rather than fixed constants, the system can explain why different observers might interpret the same map data differently—seeing either a single meandering river or a series of lakes connected by channels.
While the current prototype successfully demonstrates automated classification, future work remains in refining metrics for size and improving the logic for aggregating individual "stretches" into complex, unified geographic objects. This research represents a significant step toward GIS systems that can reason about the world with the same nuance as human language.
Summary updated