Geospatial Reasoning with Shapefiles for Supporting Policy Decisions

Policies are commonly defined as decision-making assets that express one or more actions allowed or recommended under certain conditions. In the radio communications domain, policies are created to help manage the use of a limited electromagnetic spectrum. Many policies are location-specific, meaning that they are only applicable when the usage of the radio spectrum is to occur in specific geographic locations, as dictated by the policy. In the United States, the National Telecommunications and Information Administration Manual of Regulations and Procedures for Federal Radio Frequency Management¹¹1http://bit.ly/NTIA_Redbook (NTIA Redbook) is a compilation of regulatory policies that define the conditions organizations, systems, and devices must satisfy to compatibly share radio spectrum while minimizing interference. Because policies in the NTIA Redbook regulate both commercial and federal spectrum usage, it is common to find military facilities, as well as regulations covering domestic and international locations.

The US Census Bureau publishes geospatial datasets about the United States, its territories, and points of interest, in its Census.gov data portal. The datasets contain high-definition polygons, usually in the shapefile [4] format, of many locations referred by radio spectrum policies. Although this format is a popular choice for encoding geographical entities, its main use-case is to support data interchange among geographic information systems (GIS). The shapefile format usually requires the use of a GIS to allow operations over the data, including calculations and queries. Therefore, it is not very suitable for integrating with ontology-based applications.

We present an approach to allow ontology-based applications to leverage geospatial data in formats not easily accessible or referred from within ontology constructs, and to use this data to perform geospatial calculations. We implemented the approach to support automated radio spectrum policy decisions. This is accomplished by representing Census.gov relevant polygons in the GeoSPARQL [5] vocabulary, and by defining OWL [9] classes that encode the policies’ location rules. A policy evaluation pipeline that mixes OWL reasoning and GeoSPARQL leverages this model to elicit spatial relationships, providing high-definition spatial calculations. This approach was evaluated to perform well in terms of coverage of geospatial requirements, as elicited by domain experts. It is implemented as part of the Dynamic Spectrum Access (DSA) Policy Framework [14], which was developed to serve as a machine-readable policy repository to support increased automation of policy evaluations.

The majority of location-sensitive policies in the NTIA Redbook refer to these types of locations: military facilities, States, or Country. Because the policies are originally authored in natural language, and targeted at spectrum managers, they refer to locations by their names (e.g. “Fairbanks”, “Camp Parks”), without a comprehensive definition of the boundaries of such regions. To support automation of policy decisions, it becomes crucial to encode and leverage the polygons for these relevant locations.

The Census Bureau is the United States agency that serves as the nation’s leading provider of quality data about its people and economy. Yearly, the agency publishes updated and authoritative geospatial datasets to provide meaning and context to statistical data the bureau produces. Largely published in the shapefile format, the published data²²2https://www.census.gov/programs-surveys/geography/geographies/mapping-files.html does include State boundaries and military installations, conveniently supporting our policy use-case. The State dataset is composed of 56 polygons, representing the 50 U.S. States, District of Columbia, plus 5 U.S. territories. The military installation dataset has 859 polygons, describing information about airports, laboratories, training areas, etc. In addition to the polygons, the datasets contain some minimal metadata about the locations, including a unique ID, and a legal name.

We have applied the Semantic Extract, Transform, and Load-r (SETLr [12]) to these datasets. SETLr orchestrates ETL pipelines by the use of a script in Turtle format that defines data sources, extract and transform processes, and destination formats. SETLr was executed in both geospatial datasets to extract shapes’ information and transform them into geographical features using the PROV-O [3] (prov:Location) and GeoSPARQL (geo:Feature, sf:Geometry) ontologies. Because each phase of the ETL pipeline in SETLr is defined as an RDF resource, the complete provenance of how these geographical features came to be is maintained, thereby supporting the explanation of policy decisions in more complex scenarios where multiple locations sources are involved.

Geospatial reasoning is a crucial capability when evaluating policies. Many policies, including those that regulate radio spectrum usage, are only applicable when their specified location rules are satisfied. These locations include named locations that can be mapped to features from geospatial datasets, and polygons defined directly in the policy’s rules. Either way, location rules need to be correctly evaluated, taking into consideration which polygons the policy regulates, as well as coordinates that are subject to evaluation (e.g. where a radio transmission is to occur).

We designed the DSA Policy Framework [14] to serve as a machine-readable, radio spectrum policy repository that can be used to automatically process radio transmission requests. The framework utilizes the World Wide Web Consortium’s (W3C) OWL 2 and PROV-O, and the Open Geospatial Consortium’s (OGC) GeoSPARQL 1.0 standards as a modeling foundation of radio spectrum policies and involved entities. Figure 1 shows the RDF model of a transmission request within the DSA Policy Framework. Transmission requests are defined as prov:Activity, with the associated requester as a prov:Agent. Attributes that further characterize the transmission are represented using either PROV-O (including the location attribute) or a domain ontology. Coordinates in which requesters are located are represented as Well-Known Text (WKT) [2] string, and expressed using the geo:asWKT predicate.

To allow the evaluation of the relationships between coordinates from transmissions and policy-regulated locations, we have pursued the representation of these locations as an OWL ontology where classes represent policy locations. To exemplify this approach, we will use the second provision of the US91 policy from the NTIA Redbook, which reads (with adaptations):

“In the sub-band 1761-1780 MHz, Federal earth stations in the space operation service may transmit at the following 25 sites and non-Federal base stations must accept harmful interference caused by the operation of these earth stations: Fairbanks, Camp Parks, … .”

Besides the policy text itself, which explicitly lists 25 sites where the policy is applicable, US91 is listed in the NTIA Redbook under the United States table, because it is only applicable in the US and not internationally. Listing 1 shows the representation of the involved locations for supporting this policy. Lines 1-7 define the USLocation class for expressing the entire United States land. This class is defined as a prov:Location and is a union of all States, District of Columbia, and territories from the appropriate Census.gov dataset, using the geo:sfWithin predicate from GeoSPARQL. Similarly, lines 9-16 extend this class to express the specific locations the above policy regulates, this time using features from the military facilities Census.gov dataset.

We worked with radio spectrum domain experts to elicit a set of geographical requirements that a machine-readable policy model needs to support. They appear in bold in the first column of Table 1. The table contains columns for Policy Representation and Request Evaluation. “Yes” indicates that the policy construct is either Relevant or it has been fully addressed and Implemented. “Partial” indicates that the current implementation meets a simplified requirement.

Most policies refer to locations by names or by coordinates (points, polygons, and circles), but sometimes a location is expressed in relation to another location. Currently, relative locations have been constrained to the ones expressed using the distance relationship. Geographical rules are defined in terms of the requester being in a location or a list of locations. Our approach implements these constructs using the geo:sfWithin predicate and OWL unions.

The works in [11, 13] proposed approaches for converting geospatial content to RDF, using mapping languages and ETL pipelines. The work in [6] allows the access of geospatial datasets, including shapefiles, using an ontology-based data access approach. Our conversion relied on SETLr, which enables the data conversion of geospatial data to RDF similar to the first two approaches, but also allows the maintenance of data transformation provenance. This maintenance is important in this use-case for supporting the explanation of policy decisions.

XACML 3.0, the eXtensible Access Control Markup Language [1], is a well-known policy language and de facto standard for representing attribute-based access control (ABAC) [10] policies and requests. Importantly, XACML provides a reference architecture for centralizing access control and a process model for evaluating requests against existing policies that inform the design of access control systems across domains and technologies. Thi [15] proposes an OWL-based extension to XACML to support a generalized, context-aware, role-based access control (RBAC) model, providing Spatio-temporal restrictions and conforming with the NIST RBAC standard [8]. Their work augments the XACML architecture with new functions and data types.

Our approach combines OWL, PROV-O, and GeoSPARQL to encode geospatial features, and an OWL reasoner to realize location class memberships. Our representation builds on previous work by matching the cross-domain policy expression semantics of XACML, extending it with the capacity to express rich Spatio-temporal restrictions, enabling the implementation of a wide variety of attribute-based policies across domains.

This paper presents an approach for leveraging geographical features, originally in shapefiles, to support policy decisions. In the radio spectrum domain, it is commonplace for policies to regulate the usage of the spectrum in specific locations, therefore requiring spatial reasoning to identify relationships between radio transmitters’ coordinates and policy-regulated regions. This approach is an integral part of the DSA Policy Framework, which is functioning as a prototype policy management system in support of spectrum sharing operations.

Future work involves the research and development of the application of more spatial relationships, including relative locations. Besides, in other policy publications, we have encountered locations that are expressed in unusual shapes. These include paths, cones, and altitudes. More research is necessary to assess the impact in both modeling and reasoning, should we pursue this line of work. Finally, we are generalizing the approach to beyond radio spectrum policies by initially supporting practitioners from multiple domains in the creation of policies utilizing terminology and entities in domain knowledge graphs [7].