Spatial Uncertainty for Features & Functions (SUFF) Model

This is a Semantic Web model that builds on elements of the GeoSPARQL 1.1 ontology to represent uncertainty in the location of geometric features, supporting cartographic representations as well as analytical functions.

This model uses the GeoSPARQL 1.1 namespace and those of several others from well-known and mostly standardised ontologies as well as its own - suff. All namespaces used, and prefixes for them, are listed in the table below.

This model’s specification conforms to the OntPub Profile which is a standard for ontology publication that mandates certain structural and metadata properties for the model as a whole and model elements.

Representing uncertainty in location is one of the most longstanding—and tricky—topics in geographic information science. Many practical applications have the need to represent features with uncertain locations, including representing:

These three examples relate to the three fundamental types of "imperfection" (uncertainty) in spatial data defined in WORBOYS-CLEMENTINI: inaccuracy (errors or a lack of correctness in data); imprecision (a lack of detail/exactness in data); and vagueness (a special case of imprecision, where precise boundaries between concepts remain ill-defined). Many other subtypes of uncertainty (such as bias, completeness, ambiguity, granularity, see DUCKHAM-BOOK) can be defined in terms of these three fundamental types, summarized in the figure below.

GeoSPARQL defines spatial Features and Geometry classes that can be used, with elements from other well-known Semantic Web models, for data with uncertainty in all the above scenarios. However, without dedicated methods for this, the representation and handling of it will differ in different implementations.

Consequently, this model provides:

Further, the model provides a Geometry Collection class which can be used to associate multiple geometries with a feature. This is the core of this model since multiple geometries for a feature can be used to convey location uncertainty in several ways - see the Principles below.

GeoSPARQL 1.1 does not, however, provide all the elements needed for deterministic representations of feature location uncertainty, nor does GeoSPARQL 1.1 describe how to perform topological functions on features with uncertain location representation. This model defines conventions for GeoSPARQL 1.1 element use, and introduces several new model elements, for this purpose.

Where spatial uncertainty is handled in mainstream non-Semantic Web GIS today, such as ESRI’s ArcGIS^[1], it is generally handled only cartographically and then only superficially - in the display of the data, not encoded within the data. This leads to the issue of reproducibility, where cartographic representations are not precisely defined, persistently stored, or reproducible on other systems.

By contrast, this model encodes uncertainty in the data itself. Together with the rules presented in the Principles section, the model can provide a basis for any software system to display uncertainty cartographically or analyse uncertainty spatially (such as probabilstic topological interactions).

This model only directly addresses the representation of uncertainty in the location of spatial features, where that uncertainty can be captured using a set of geometries.

The model does not address uncertainty in the values of other attributes or properties of features. Other ontologies do exist to provide for the representation of measurement uncertainty, including simple models such as schema.org include properties such as marginOfError for this purpose.

Further, this model is not intended to be used to represent uncertainty in location when a simple property associated with a single geometry would suffice. Such simple representations are already accessible through elements from GeoSPARQL, see:

Uncertainty in the location of a feature is captured by linking the feature not to one geometry but to multiple - a geometry collection. Each geometry in the collection conveys part of the representation of the feature’s location together with a value either being given in the data or inferrable due to the geometry’s relative position in the collection.

The interpretation of the value is partly captured through three levels of measurement, as follows.

The GeoSPARQL standard provides both an ontology which defines spatial elements and predicates to indicate topological relationships between them and also function definitions that can be used to calculate those predicates.

For example, the predicate geo:sfWithin indicated that its subject is spatially within its object where within is the Simple Features Within relation defined by OGC-06-103r4. The related GeoSPARQL function is geof:sfWithin which defines the parameters and results from the function used to calculate the `geo:sfWithin relation between two spatial objects.

One objective of this model is to enable the definition of rules that propagate uncertainty in location through to levels of certainty in topological relation. There are an enormous range of different possibilities when computing such a relation in the literature, including COHN-GOTTS, CLEMENTINI-FELICE, BJORKE, WINTER, ROY-STELL to name but a few.

This model uses a vocabulary of Levels of Measurement to provide values for the range of the predicate level of measurement. that vocabulary is published online at https://linked.data.gov.au/def/levels-of-measurement.

A simple listing of that vocabulary is also given below.