Automatic Detection of Ports For Map Generalisation William Mackaness Phil Bartie The School of GeoSciences, Biological and Environmental Sciences The University of Edinburgh, University of Stirling Edinburgh EH8 9XP Stirling FK9 4LA william.mackaness@ed.ac.uk phil.bartie@stir.ac.uk Abstract 1 The vision of automated cartography is built around the idea of a highly detailed representation of the world from which we can construct a continuum of dynamic outputs ranging to the very smallest scale. Deriving higher order objects from the grouping of lower order constituent parts can be framed as a pattern recognition task. We can devise a prototypical representation of a higher order object – represented as an ontology. The ontology can then be used as a basis for searching the database for higher order objects (eg cities, airports, ports) based on the proximity between their constituent parts. A city defined in terms of {suburbs, municipal buildings, transport networks}; airports in terms of {runways, passenger terminals, taxiways}, and ports in terms of {harbours, docks, container ports}). Such an approach provides a basis for extending the range of scale dependent representations of geographical concepts – in turn facilitating the automated creation of thematic maps. By taking account of the geographical composition of such higher order objects we can extend automated map generalisation over larger changes in scale. Such thinking chimes with ideas in Gestalt theory (in reaction to structuralist approaches) in which the geographical concept emerges from the juxtaposition of a particular set of finer scale concepts. Introduction Much research in map generalisation is structuralist in nature; that is to say it breaks the problem down into distinct and unrelated elements that unfortunately often extinguish the notion of geography. The considerable and sustained focus around the Douglas Peuker Algorithm is an excellent example of structuralist thinking with all its associated problems and limited application domain. As an antidote to such thinking, increasingly researchers are exploring ways of making explicit the metric and topological properties between collection of entities in order to construct higher order entities (for example that the concept of a city can be considered to be made up a dense collection of transport networks, municipal and industrial buildings and residential areas). This idea of automatically detecting entities from their ‘sub entities’ has been explored by (among others) Chaudhry et al (2009) in which they demonstrated the automatic detection of schools, retail parks and airports (defined through the detection of their functional elements). Thus ‘playing fields’, ‘car parks’, ‘classrooms’, ‘sports facilities’ might variously define the extent of a School and from a map generalisation perspective, we can envisage replacement of these finer scale objects with a single polygon (or its centroid) at smaller scales. Such an approach worked well for entities composed of contiguous objects (eg airports, retail parks, railway stations) but some higher order objects are much more complex – in their composition (how they are constituted), their geographical extent, and how they ‘interact’ with other phenomena. An intriguing example is the entity ‘Port’. 1 Mackaness, W.A. & Bartie, P. 2016 Automatic Detection of Ports For Map Generalisation, Workshop of the ICA Commission on Generalisation and Multiple Representation ‘Automated generalisation for on-demand mapping’, Helsinki, Finland, 14 June, 2016