“Geography matters, not for
the simplistic and overly used reason that everything happens in space, but
because where things happen is
critical to knowing how and why they happen.”
Barney Warf and Santa Arias: Introduction to The Spatial Turn
My research questions center on the role of space in information science. I study how to ask and answer spatio-temporal questions, testing new ideas by designing application programming interfaces (API) and spatio-temporally enabled knowledge infrastructures.
Today’s technologies for spatial computing, notably Geographic Information Systems (GIS), are organized by methods to answer spatial questions. What language should a domain expert use to ask spatial questions? My core concepts of spatial information provide vocabulary for such a language. They are in the process of being formalized and tested for a broad range of applications, connecting user questions to GIS and related technologies.
We do not know how to make hypotheses spatially explicit. What pattern underlies their formulation? How did Dr. Snow move from the idea that cholera might be transmitted by drinking water to mapping patients and water pumps? Scientific theories make predictions about processes, whose participants (such as cholera patients and water pumps) can be located. A method to spatialize hypotheses will therefore require studying how process models relate participants in space. While not all participants of processes in science can be located in space, many can, and in some cases their location matters to how and why the processes occur (see the above motto).
What does an observation capture? Sensors and crowd-sourcing supply big observation data in need of interpretation. My Functional Ontology of Observation and Measurement (FOOM), as well as the Semantic Sensor Network ontology of W3C derived from it and other observation ontologies, need to be extended to deal with the positioning of observations and with their resolution, as well as to support event recognition. The resulting shared vocabularies will help domain scientists to annotate, expose, link, query, and interpret observation data.
The handling of spatially referenced data from their collection through cleaning and documenting to exposure and access at various portals is often haphazard, preventing smooth scientific workflows. Typical problems include a lack of awareness of existing data, difficulties to find data, obscure formats, lack of documentation through metadata, unspecified units, inappropriate null values, ambiguous semantics, changing locations of repositories, inhomogeneous and unclear quality, etc. How do semantic web technologies help scientists to deal with their own and other people’s scientific data more easily?
Spatial reference systems solve the problem of referring horizontal positions anywhere on earth to coordinate systems and of mediating between multiple systems; gazetteers achieve the same for place names; temporal referencing is solved for current practical purposes in geographic space. What does it take to make attribute or thematic referencing and mediation equally usable? A general theory of grounding measurement scales, anchoring the digital in the physical, is the next step in operationalizing semantic referencing. Furthermore, stochastic mediation methods need to be developed to avoid always having to “go to the ground”.
The quantitative spatial referencing of coordinate systems and the place name referencing of gazetteers fail to deal with qualitative spatial descriptions (“on your right past the second traffic light”). Qualitative spatial reasoning still lacks a general theory for generating and interpreting such descriptions. The available methods are based on informal classifications of reference frames, typically limited to directional information, rarely use qualitative calculi, and cannot deal with multiple levels of granularity. Fixing some of these shortcomings will help progress toward a general theory of spatial referencing and to design adequate shared vocabularies for place descriptions.
The legacy of GIS comes with a wide range of spatial data types, making GIS hard to learn and use. Which of these data types are really needed for spatial computing and which are just convenient (and sometimes inconvenient) extras? The hypothesis explored here is that all spatial computing can be done with just two spatial data types, grids and graphs. Grids are regular subdivisions of a space (“raster data”); graphs are topological structures with nodes, edges, and faces. Point sets can be considered as degenerate graphs. The main advantage of the proposed conceptual reduction would be to base spatial computing entirely on the mathematically well-founded techniques of image processing and network analysis.
The proliferation of GIS feature types and their ontologies turns semantic interoperability into a daunting challenge. But are we solving real problems when distinguishing rivers from streams or roads from streets? It is often more important to capture semantics at the next level of abstraction, for example in terms of what can move on water or on land. Today’s ontology languages cannot express these distinctions easily, focusing instead on fine-grained distinctions at the type level. Big data turn this undertaking into a Sisyphus task, but recent work on ontology design patterns shows how to step up the abstraction ladder. Haskell, the modern standard functional language, provides the necessary formal modeling capabilities through its multi-parameter type classes. The goal of this research is to connect existing ontologies and database schemas at the type level to ontology design patterns derived from type classes.