We present a software tool for urban planing, driven by semantic web ontologies. With this tool, the participants of the urban development process can semantically annotate an intervention site with knowledge about the use of the land. The tool provides the user as much exibility as possible in the choice of the land use standard that will be used to perform the classi cations. This exibility is achieved by de ning extensible ontologies in which the concrete standards can be built upon. This allows our tool to load external ontologies that de ne the taxonomy of the standard and the semantic relations between the land use categories. The loaded ontologies seamlessly integrate in the application, and are used through a friendly user interface. We also demonstrate how ontologies can be used to perform automatic categorizations and consistency checks in the geographical knowledge, aiding in the process of identifying the land uses of a site.
Urban and regional planners usually develop long and short term plans to guarantee e cient management of land use and to promote the growth and revitalisation of urban, suburban, and rural communities. Before making plans for the development of these communities, planners study and report on the existing land use for business, residential, and community purposes. Our work aims to design and build a tool for the manipulation of land use knowledge, to assist the urban development process. It is targeted at district planning and the ultimate goal is to promote the generation of more sustainable urban environments. To achieve this goal is important to have reliable and up-to-date georeferenced information, rstly, in order to identify the requirements of the plan, and subsequently to establish appropriate recommendations.
We have designed 4CitySemantics, a software tool that aims at aiding in the city planning process, and heavily relies on semantic web technologies. The main objective of this tool is to provide the urban planners with a sharable semantic interpretation of a de ned intervention site and surrounding areas. Besides visualizing the semantic information associated with georeferenced spatial features, this tool provides the participants in the urban design process with the possibility to identify and classify semantically the urban areas subject to intervention.
The key principle behind the tool is to o er as much exibility and customization as possible, in order to be able to integrate smoothly with di erent urban knowledge and di erent classi cation standards. Allowing the tool to work with any land use classi cation standard is paramount because there are many classi cation systems used in land use planning. Land use standards are usually de ned by state regulations, and the urban planner is required to use the standards existing in the intervention zone. Semantic web technologies play the main role for attaining the desired exibility. We de ned base ontologies written in OWL 2, which can be extended as needed to describe the standards that the user wants in the tool. The tool will then interpret the knowledge present in those extensions using the semantic structure laid out by the base ontologies.
With 4CitySemantics users are able to 1) load a georeferenced map from a shape le, 2) choose the intervention zone and de ne the surrounding region on the map; 3) de ne any other zones on the map as necessary; 4) load a land use standard from an OWL ontology; 5) visualize the standard and all the information associated with it; 6) classify the parcels on the map in accordance with that standard; and 7) query the map for the parcels classi ed with any land use categories, taking advantage of the reasoning capabilities. 2
The use of ontologies has been a hot topic in urban planning [
In the context of Towntology, CityGML [
KnowledgeScapes [
In 4CitySemantics, knowledge regarding geographical features and land use classi cation taxonomy are represented using OWL 2. The knowledge is separated from the tool, and so it is fully independent of any land use standard, as long as it is represented as an ontology. Designing a system where the user can dene its own concepts, which our tool should be able to correctly interpret them, raises some challenges: In one hand, it must be given enough liberty for the user to de ne its own concepts, allowing rich descriptions and fully supporting the OWL 2 semantics. On the other hand, the application must be able to bring into play these foreign concepts, imposing some (preferably minimal) restrictions to the user in the construction of the ontology.
The compromise between these goals is achieved through a small set of concepts that are de ned in non-modi able ontologies in which the tool relies. We call these ontologies the skeleton ontologies, because they are the common ground where di erent ontologies will be built upon. To create an extension ontology de ning a concrete land use, the user must extend the concepts of the LandUseSkeleton ontology, with all the expressiveness he wants, as long as the connection points between the ontologies are kept, so the tool can interpret it. The land use skeleton ontology can be viewed as a schema for the de nition of a land use, which itself is also a schema. The extension provided by the user together with the skeletons provided by the application form the T-Box of the knowledge base. They de ne the schema that is used to assert information about the parcels of the map.
The semantic information will target geographical features from a map (loaded via a shape le), so there must be some correspondence between the map and the ontology individuals. This is speci ed in the CoreGIS ontology, that de nes the classes Feature and Zone and some properties related to them. A geographical feature is any element in the map, which may be a polygon, line or point. A zone is an aggregate of geographical features. For each geographical feature present in the map, there will be a corresponding OWL individual of the class Feature, linked to the map by the identi er of the feature. The ontologies are not intended to replace the shape le, just complement it with semantic information, so it is not necessary to express the features geometry in the ontologies.
The semantic information created in the tool is stored in a separate OWL ontology respective to the current project. This ontology is representative of the A-Box. It can be used to 1) perform future modi cations and additions to the information by opening the ontology in the tool 2) share the information, exploiting the advantages o ered by the Semantic Web. The articulation of all the ontologies described in this section is summed up in Figure 1.
The tool provides a friendly user interface for the loaded land use ontology and the georeferenced map. The user can annotate semantic information about the map parcels without worrying about the implementation details of the ontologies. If it is already built an ontology representing the desired land use, the participants of the urban planning process can use the tool without even understanding OWL { and simultaneously take advantage of it.
T-Box Land Use Concretization LandUseSkeleton.owl
CoreGIS.owl
A-Box
Project Ontology
4CitySemantics is developed in Java 7 over Netbeans Rich Client Platform 1. The ontologies are read using the OWL API2, and the reasoning throughout the application (which will be detailed later) is provided by the Pellet Reasoner3. The ontologies were built using Protege 4.14. The georeferenced maps are loaded and manipulated in the application with GeoTools5. 4
In this section we describe the skeleton ontology that allows any land use
standard to be used in the application. An extension of this skeleton will be
exemplied with the LBCS6 ontology, that was developed in previous work [
Through the study of multiple land use standards we identi ed a few common characteristics that formed the skeleton ontology, which is the common ground where the actual land uses will be built upon. Namely, we found that each land use category has an identi er (code), a color for visual identi cation in maps, a title and a description. To cope with these necessities, it was created a LandUseCategory class and data properties for the color (hasColor) and identi er (hasIndentifier), both with String as range.
The land use standards typically show some degree of hierarchic organization, so the land use categories should be represented as classes in order to map the 1 Generic framework for Swing applications { http://platform.netbeans.org 2 Java API for manipulating OWL Ontologies { http://owlapi.sourceforge.net 3 An OWL 2 DL reasoner { http://clarkparsia.com/pellet 4 Open source ontology editor supporting OWL 2 { http://protege.stanford.edu 5 Open-source Java GIS Toolkit { http://geotools.org 6 Land Based Classi cation Standards { http://www.planning.org/lbcs hierarchy of land use categories. To assert information about the categories, namely through the hasColor and hasIdentifier properties, it should be used punning. Punning7 is a form of meta-modeling supported by OWL 2 that makes possible to use a class in places where an individual was expected (among other possibilities). This is achieved in the ontology by declaring an individual with the same IRI (name) as the class that is being punned. While the object shares the same IRI, depending on its context, it is evaluated by an OWL reasoner as a di erent thing (class or individual).
To create an ontology describing a land use standard, the user has to import the skeleton ontology, create subclasses of the LandUseCategory class, associate them with proper annotations for label and description, and ll the color and identi er properties using punning in the classes. We can see an example of this in Figure 2. Additionally, if there is already an ontology describing a land use standard, it can be linked with the skeleton, by declaring the existing categories as subclasses of LandUseCategory and de ning relations between the existing properties and the ones from the skeleton (e.g. equivalent properties).
To show the land use taxonomy in the user interface (Figure 3), the application retrieves the subclasses of LandUseCategory and the respective properties in the punned individuals. The categories are displayed with all the information available. The user can select any of these land use categories and associate them with geographical features through a friendly user interface. The geographical features are associated to land use categories by the object property hasLandUseCategory (Feature ! LandUseCategory). When the user classi es some features, the expression hasLandUseCategory some C, where C is the selected category, is added as a type to the corresponding OWL instances of the geographical features selected in the map. 4.2
The most popular GIS software tools, such as GRASS GIS8 or Quantum GIS9, rely on relational databases to represent the information associated with the map parcels, including the land use. In these tools, the information is not semantically related, and so the user has the burden of, for example, maintain the integrity of the information, or develop more complex queries to deal with the relations between the categories. Our tool takes advantage of reasoning over the ontologies and can overcome some of these di culties. It is important to note that the direct representation of the land use taxonomy (in the previous section) is only the tip of the iceberg. The land use ontology can also contain axioms de ning any other relations between the concepts. These axioms will be used by the reasoner and will a ect the actions of the user in the tool. This raises interesting possibilities in this domain, some of which will be presented next. 7 Metamodeling in Domain Ontologies { http://techwiki.openstructs.org/index.
php/Metamodeling_in_Domain_Ontologies 8 GRASS GIS { http://grass.fbk.eu 9 Quantum GIS { http://www.qgis.org Fig. 3. A screenshot of the tool. One land use category is selected (left), displaying in the map (right) the features with that category. It shows some features selected in the map (in yellow), and the user classifying them with the selected category. For the selected category, it is shown the respective color, full name and description (left). Finding the features with the selected category The most obvious use of reasoning is to nd which features are directly or indirectly categorized with a given category. This is achieved by asking the reasoner for the instances of hasLandUseCategory some C, where C is the category. Note that this scheme deals perfectly with hierarchically organized land use categories, for example if we asked for Educational Services, we would also obtain features categorized with College, High School, etc. The features belonging to the selected category are then coloured in the map according to the hasColor property of its category.
It is also possible to conceive more general queries, such as asking for all the features that have a de ned Function (using the example of LBCS, Figure 2). This highlights in the map features with many di erent and disjoints sub-categories. Each of these features is coloured according to the color of its sub-category, resulting in a multi-colour map, useful for land use analysis. Automatic categorization (inference) Automatic categorization of some geographical features can be achieved through the use of equivalent classes axioms. We could for example state that all the features that have both the category ResidenceFunction and PrivateOwnership should be categorized (are equivalent to) as PrivateResidential. These equivalences can be anything else, including universal and quanti er restrictions, and expressions that are not related to the land use standard itself, but some other property of the geographical feature. This scheme can also be used to make correspondences between categories of di erent land use systems.
Consistency checks Through the creation of some equivalent (de ned) classes and disjoint axioms it is also possible to express some restrictions and incompatibilities that will allow a more robust and error free classi cation. For example, we can express that a School cannot also be a Hospital by specifying that the classes SchoolFeature (equivalent to hasLandUseCategory some School) and HospitalFeature (equivalent to hasLandUseCategory some Hospital) are disjoint. Whenever the user classi es geographical features in the application, the consistency is checked by the reasoner. If the ontology becomes inconsistent, that classi cation is immediately removed in order to revert the ontology to a consistent state. The features that were the cause of the inconsistency are also highlighted on the map, so the user can understand its error. 5
We have presented 4CitySemantics, a tool for urban planing based on semantic web ontologies. Using the tool, the participants in the urban development process may semantically annotate the intervention site with knowledge about the use of the land. We show that ontologies can be useful when dealing with this type of information and can o er some advantages over the more traditional tools, for example the inference and consistency checks o ered by the reasoning.
We have shown how extensible OWL ontologies can be used to create a exible application, which adapts its behaviour based on an ontology that is provided by the user. Despite using external ontologies, the application can integrate seamlessly with them, and provides a user interface that hides the implementation details of the ontologies. Following this principle, 4CitySemantics was designed to be independent of urban semantic standards, being easily adaptable to di erent standards.
The development of the tool was followed closely by urban planning architects, the target of our application, and people familiar with geographic information systems. It was also presented to people outside of the project. The users considered the tool very interesting, and stressed that the possibility of using any land use standard was important. Despite some initial scepticism from the users familiar with the traditional GIS tools, they were pleased by the reasoning capabilities o ered in 4CitySemantics. It could e ectively tackle problems they had when working with the traditional tools, for example when dealing with hierarchical land use taxonomies.
Acknowledgements This research originates in the City Induction project supported by Fundac~ao para a Ci^encia e Tecnologia (FCT), Portugal, hosted by ICIST at the Technical University of Lisbon (PTDC/AUR/64384/2006) and co-ordinated by Professor Jose Pinto Duarte. N. Montenegro is funded by FCT with grant SFRH/BD/45520/2008.