Exploiting empirical knowledge for automatic delineation of city centres from large-scale topographic databases

Current topographic databases rarely represent higher order geographic phenomena, such as city centres. However, such concepts are often referred to by humans and used in various forms of spatial analysis. Hence, the value and usability of topographic databases can greatly be improved by methods that automatically create such higher order phenomena through cartographic pattern recognition techniques, departing from the very detailed, geometry-oriented representations of topographic databases. As many higher order phenomena are only vaguely defined, this paper develops and evaluates a methodology to acquire definitional knowledge about geographic phenomena by participant experiments and use this knowledge to drive the cartographic pattern recognition process. The method is applied to acquire knowledge about British city centres and delineate referents of city centre from topographic data. City centres produced for ten British cities are compared to areas derived from alternative sources. F1-scores between 0.45 and 0.88 are achieved, suggesting that the delineation produced plausible city centre areas. The benefits of our work are better (and user-driven) descriptions of complex geographic phenomena that can form the basis for accurately enriching topographic databases with additional semantics, thus yielding added value for the data producer and the end user.

Figure optionsDownload as PowerPoint slideHighlights
► Comprehensive methodology including semantic modelling, pattern recognition and evaluation.
► Participant experiment to acquire definitional knowledge of UK city centres.
► Computational implementation derives city centre typicality from topographic data.
► High plausibility of generated city centre boundaries by comparison to other sources.
► Good correspondence of computed typicality values to human assessment of images.