Ontological Representation of Constraints for Geographical Reasoning

Gianluca Torta

, Liliana Ardissono

, Marco Corona

, Luigi La Riccia

and Angioletta Voghera

Dipartimento di Informatica, Universit

a di Torino, Torino, Italy

Dipartimento Interateneo di Scienze, Progetto e Politiche del Territorio, Torino, Italy

Keywords:

Geographic Knowledge, Geographical Constraints, GeoSPARQL, Ecological Networks, Urban Planning.

Abstract:

We describe a framework that supports multiple types of constraint-based reasoning tasks on a geographic

domain, by exploiting a semantic representation of the domain itself and of its constraints. Our approach is

based on an abstract graph representation of a geographical area and of its relevant properties, for performing

the reasoning tasks.

As a test-bed, we consider the domain of Ecological Networks (ENs), which describe the structure of existing

real ecosystems and help planning their expansion, conservation and improvement by introducing constraints

on land use. While some previous work has been done about supporting the veriﬁcation of compliance of fully

speciﬁed ENs, we aim at taking a signiﬁcant step further, by addressing the automatic suggestion of suitable

aggregations of land patches into elements of the EN. This automated generation of EN elements is relevant

to support the human planner in the design of public policies for land use because it leverages automated tools

to carry out a possibly lengthy and error-prone task.

1 INTRODUCTION

This paper describes a framework supporting

constraint-based reasoning tasks on a geographic

domain. We take the motivating example of Eco-

logical Networks (ENs) (Bennett and Mulongoy,

2006) as a starting point to deﬁne a more general

geographic reasoning model in which both the

domain and its constraints are uniformly represented

as OWL ontologies (W3C, 2017). In this way, we

can beneﬁt from the expressiveness provided by

standard Semantic Web languages in the speciﬁcation

of domain knowledge.

ENs have been introduced to describe the structure

of existing real ecosystems and help planning their

expansion, conservation and improvement by impo-

sing constraints on land use. However, up to now,

the design of the ENs structure, and the deﬁnition of

public policies proposing land use transformations to

comply with the EN constraints, have been carried out

manually, in a lengthy and possibly error-prone type

of activity. This is due to the informal speciﬁcation

of ENs, which consists of a rather large number of

guidelines expressed in Natural Language.

In this paper, we propose to represent both the EN

domain and its constraints in a semantic knowledge

representation language, with the aim of supporting a

number of automated reasoning tasks concerning the

design of the structure of the EN, as well as urban

and regional plans and projects for transforming a ge-

ographic area so that it (better) complies with the EN

speciﬁcation. A desired side effect of those activities

is to construct the social awareness on bindings and

opportunities related to ENs for quality of life.

Speciﬁcally, we aim at developing a framework

for the automated suggestion of suitable aggregations

of land patches of a geographic area into elements of

an EN. The main contributions of this paper are:

• An ontological representation of ENs and of the

related constraints on land use which supports

knowledge sharing and semantic reasoning. Our

work fully adheres to the GeoSpatial Semantic

Web paradigm (Janowicz et al., 2012a), which

promotes standard knowledge representation lan-

guages to maximize data and application intero-

perability.

• A model for the automated generation of a graph-

based, abstract representation of a geographical

area, which speciﬁes the ecological and land use

characteristics of land patches, and their adja-

cency relations. Speciﬁcally, the nodes of the

graph represent homogeneous pieces of land, each

one associated with a given set of deﬁned proper-

ties; e.g., naturalness, or irreversibility of the land

use. The arcs represent abstractions of the adja-

136

Torta, G., Ardissono, L., Corona, M., Riccia, L. and Voghera, A.

Ontological Representation of Constraints for Geographical Reasoning.

DOI: 10.5220/0006960901360147

In Proceedings of the 10th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2018) - Volume 2: KEOD, pages 136-147

ISBN: 978-989-758-330-8

cency relations between nodes.

• An extensible framework that exploits the graph-

based model for performing reasoning tasks on

the EN domain, guided by the knowledge enco-

ded in both the domain and constraints ontologies.

As a proof-of-concept, the current implementa-

tion of our framework is equipped with two rea-

soners optimized to solve speciﬁc types of tasks:

i.e., ﬁnding paths that connect certain areas, and

clustering areas characterized by the same given

properties.

Even though the ultimate goal of our work is to pro-

vide ontologies and specialized reasoners that allow a

fully-automated, mixed-initiative management of re-

asoning tasks on geographical areas, in this paper, we

focus on tasks relevant to the ENs domain.

The remainder of this paper is organized as fol-

lows: Section 2 provides some background on seman-

tic knowledge representation and ENs. Sections 3 and

4 present our knowledge representation and reasoning

model. Section 5 describes the framework implemen-

tation and Section 6 positions it in the related litera-

ture. Sections 7 and 8 present our future work and

conclude the paper, respectively.

2 BACKGROUND

2.1 Semantic Representation of

Geographic Information

The GeoSpatial Semantic Web vision, presented in

(Janowicz et al., 2012a), advocates for represen-

ting geographical information by means of ontologies

suitable for explicitly describing concepts and relati-

ons among concepts. This is important to support a

conceptual notion of data interoperability, which goes

beyond the adoption of a common representation for-

mat, and is aimed at enabling correct data interpreta-

tion and inferences in geographical reasoning.

Even before the birth of this vision, several geo-

graphical ontologies were developed to support infor-

mation sharing and retrieval. For instance, Fonseca et

al. proposed an ontology to classify geographic ele-

ments with respect to geometric characteristics and at-

tribute values; i.e., semantic features (Fonseca et al.,

2002a). Moreover, they discussed semantic granula-

rity, i.e., the level of detail at which geographic ob-

jects are described, focusing on interoperability is-

sues; see (Fonseca et al., 2002b).

Some general ontologies have been published to

share toponyms and generic geographic concepts;

e.g., the GeoNames Mappings ontology (GeoNa-

mes.org, 2018) based on the GeoNames database

(http://www.geonames.org/). Researchers have also

focused on spatial granularity to describe toponyms

at different levels of detail; e.g., see (Palacio et al.,

2015). Other ontologies have been developed to pro-

vide a semantics of Volunteer Geographical Informa-

tion. For instance, LinkedGeoData links crowsour-

ced OpenStreetMap (OSM) information to GeoNa-

mes and others ontologies (Janowicz et al., 2012b).

Moreover, focusing on Open Data and crowdsourced

data, Baglatzi et al. (Baglatzi et al., 2012) bridged the

gap between ontological and crowdsourcing practices

via ontological alignment. Furthermore, OSMonto

(Codescu et al., 2011) has been deﬁned to structure

the implicit ontology of OpenStreetMap (OpenStreet-

Map Contributors, 2017) tags and to format it into an

RDF (W3C, 2017b) scheme.

Besides the work supporting geographic informa-

tion sharing and retrieval, more speciﬁc ontologies

have been deﬁned to describe ﬁne-grained aspects

of geographical objects. Of primary importance for

the present work is the GeoSPARQL ontology (OGC,

2012), which describes geographical objects suppor-

ting the speciﬁcation of their geometry, as well as to-

pologic relations among different objects.

Other ontologies relevant to the present work are

the ones that model the types of land use/cover, e.g.,

LBCS-OWL2 (Montenegro et al., 2011), HarmonISA

(Hall and Mandl, 2006). As we shall see, we have

currently adopted a taxonomy based on the Land Co-

ver Piemonte (LCP) cartography (Provincia di Torino,

2014), mainly because the experimental data availa-

ble to us was tagged according to it, see project (Citt

Metropolitana di Torino, 2014).

2.2 Ecological Networks

Ecological Networks (ENs) have been proposed to

preserve biodiversity by reducing the process of na-

ture fragmentation caused by the development of new

urbanizations, infrastructural networks and intensive

agriculture (Jongman, 1995). Recently, there has been

an exponential growth of urban land use towards more

natural spaces: external urban areas (uncultivated or

abandoned cultivated land, burnt areas, degraded fo-

rests) have often been conﬁned from urban and regio-

nal planning to an “inessential” position and someti-

mes simply considered as “waiting areas for a new ur-

banization”. As reported in (Bennett and Mulongoy,

2006), “ecological networks share two generic goals,

namely:

1. Maintaining the functioning of ecosystems as a

means of facilitating the conservation of species

Ontological Representation of Constraints for Geographical Reasoning

137

Figure 1: Ecological Network representation, from (Bennett

and Mulongoy, 2006).

and habitats.

2. Promoting the sustainable use of natural resources

in order to reduce the impacts of new urbanizati-

ons on biodiversity and/or to increase the biodi-

versity value of managed landscapes”.

Despite the Protected Areas and Natura 2000 si-

tes are now considered the backbone of European po-

licy for biodiversity, the increasing expansion of ur-

banizations and infrastructural networks is challen-

ging the conservation of natural habitats for the pre-

servation of animal species and plant varieties. The

policies for the improvement of EN are necessary to

overcome the fragmentation of the habitats and natu-

ral areas, which is the main cause of biodiversity loss

in Europe. The consequences of these processes can

be summarized in the following phenomena (Benedict

and McMahon, 2002):

1. The substantial loss of natural areas due to urban

development.

2. The fragmentation of natural areas into smaller,

disconnected patches.

3. The degradation of wetlands, which have an im-

portant ecological function for the control of wa-

ter ﬂows, the ability to block sediments, for the

support of plant and animal species, etc. (step-

ping stone function) and for the ability to provide

nutrients for the ecosystems.

4. The inability of ecosystems to respond to changes

and ﬁnd a new ecological balance: that is to say a

signiﬁcantly reduced resilience.

5. The loss of ecosystem services, such as the control

of water and the ﬁlter functions for pollutants.

6. The increased economic costs for public services,

due to the response to natural disasters deriving

from human footprint.

An Ecological Network can be deﬁned as an inter-

connected system consisting of territorial areas that

include natural and semi-natural habitats. The typical

representation of an EN is a network of core areas in-

terconnected by corridors; see Figure 1. According to

(Bennett and Mulongoy, 2006):

• Core Areas are the areas “where the conservation

of biodiversity takes primary importance, even if

the area is not legally protected”.

• Buffer Zones are the areas which “protect the net-

work from potentially damaging external inﬂuen-

ces and which are essentially transitional areas

characterized by compatible land uses”. They are

important to safeguard and increase the stability

of the core areas.

• Corridors “serve to maintain vital ecological or

environmental connections by maintaining physi-

cal (though not necessarily linear) linkages bet-

ween the core areas”.

• Sustainable-use areas are zones “where opportu-

nities are exploited within the landscape mozaic

for the sustainable use of natural resources toget-

her with maintenance of most ecosystem servi-

ces”.

So far, the work about ENs has mostly focused on the

following perspectives:

1. Ecological network analysis has dealt with stu-

dying and simulating in a mathematical way the

interaction between organisms within the ecosy-

stem, the dynamics of the relations among spe-

cies, the existence of dynamical bottlenecks in

the functioning of the ecosystems, etc.; see, e.g.,

(Fath et al., 2007; Ulanowicz, 2004; Fath et al.,

2007; Lurgi and Robertson, 2011; Gobluski et al.,

2016; Pilosof et al., 2017). This research line has

a different goal than ours. In fact, it aims at pre-

dicting the evolution of an EN starting from an

initial status, but it does not deal with represen-

ting and reasoning with the land use constraints.

2. Several ENs have been implemented at different

scales (from European down to municipality), re-

sulting in the production of different guidelines

on land use, and planning documents, expressed

in Natural Language; e.g., (Council of Europe,

2000; Bennett and Wit, 2001; Bennett and Mu-

longoy, 2006). However, their linguistic speciﬁ-

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

138

Figure 2: A portion of the EN Ontology.

cation makes it hard to present EN elements in ge-

ographical maps, which have to be manually cre-

ated. Moreover, it challenges an automated veriﬁ-

cation of EN constraints.

To the best of our knowledge, the only work pursuing

the automation of a task needed for EN project plan-

ning is the one described in (Torta et al., 2017; Torta

et al., 2018), which focuses on EN validation. As dis-

cussed later on, we extend that work by adopting a

uniﬁed semantic approach for a holistic representa-

tion of knowledge about ENs and about their con-

straints. Moreover, we adopt a more general approach

to the supported reasoning tasks.

3 KNOWLEDGE

REPRESENTATION

3.1 Ecological Networks Representation

Our EN Ontology describes the main concepts and

relations of Ecological Networks. We deﬁned it

starting from the speciﬁcations produced in project

“Experimental activity of participatory elaboration of

ecological network” (Citt

a Metropolitana di Torino,

2014), conducted by the Metropolitan City of Turin

(Italy)

in collaboration with Polytechnic of Turin

and ENEA

. Speciﬁcally, in the present work we ex-

tend and reﬁne the ontology proposed in (Torta et al.,

2017; Torta et al., 2018). The above mentioned pro-

ject aimed at deﬁning a proposal for the Ecological

Network implementation at the local level in two pi-

lot municipalities near Turin. The proposed appro-

www.cittametropolitana.torino.it/

www.polito.it

www.enea.it

ach was aimed at guiding local Public Administra-

tions with measures to limit anthropogenic land use

and, where possible, orient and qualify the conserva-

tion of ecosystem services.

Figure 2 shows the main classes of the EN On-

tology, which is developed in OWL (W3C, 2017),

and is augmented by the GeoSPARQL ontology

(OGC, 2012). The graphic notation is borrowed from

(Van Hage et al., 2011); in particular, arrows with

open arrow heads symbolize subclass relationships

between the classes, while regular arrows connect

domains and ranges of properties

. The top class of

the EN Ontology is the Feature, a class deﬁned in the

GeoSPARQL ontology and colored in dark gray for

easy recognition. A Feature has a Geometry on the

2D plane, and can thus be used to represent points, li-

nes, and areas on a geographic map. GeoSPARQL de-

ﬁnes a number of topological geometric relations be-

tween Features that correspond to fundamental relati-

ons (such as intersects, equals, contains, etc.) known

in the literature as Simple Features (Open Geospatial

Consortium et al., 2011).

In order to deﬁne ENs, we specify the Feature

class into the four hierarchies of classes representing

the core of the domain. The roots of these hierarchies

are colored in gray.

• ENElement represents an element of the EN. This

is either a Core Area (CoreArea), a Sustainable-

use area (SustArea), or a Priority Expansion Ele-

ment, which is further specialized to a Corridor

(Corridor) or Buffer Zone (Buffer).

• Intervention represents an intervention for buil-

ding, improving or conserving the EN.

• Operation represents a speciﬁc operation of elimi-

nation (Elimination), construction (NewPlanting),

The diagrams were created with the Dia tool.

Ontological Representation of Constraints for Geographical Reasoning

139

Figure 3: Portion of the LandUseElement hierarchy.

or maintenance (Maintenance) that is part of an

intervention.

• Patch represents a small area of the map characte-

rized by a speciﬁc land use; see below.

It is worth noting that a Patch P typically belongsTo

an EcologicalNetworkElement E, and E is madeOf

patches including P.

Figure 3 shows the LandUseElement hierarchy.

Each class of the hierarchy is a singleton, containing

exactly one representative object characterized by:

• A speciﬁc type of land use deﬁned in the Land

Cover Piemonte (LCP) cartography (Provincia di

Torino, 2014): e.g., wetland (WetLand), wooden

land (WoodenLand), and similar. The LCP deﬁnes

97 different types of land use at the most detailed

level (LCPlvl4) of a hierarchy which includes 45

classes at level 3 (LCPlvl3), 15 classes at level 2

(LCPlvl2), and 5 general classes at the most gene-

ral level LCPlvl1.

• Five evaluation criteria (OWL properties of Lan-

dUseElement, not shown in Figure 3) (Voghera

and La Riccia, 2019):

– naturalness (how close is the element to a natu-

ral environment);

– relevance (how relevant it is for the conserva-

tion of the habitat);

– fragility (how fragile the element is with respect

to the anthropogenic pressure);

– extroversion (how much pressure it can exert on

the neighboring patches);

– irreversibility (how difﬁcult it would be to

change the use of the element).

The value for each criterion ranges from 1 to 5,

and 1 represents the maximum value.

Each Patch of land is describedBy a LandUseEle-

ment; i.e., it is associated with a speciﬁc land use.

For our current purposes, the description of land pat-

ches with the attributes we have just described is suf-

ﬁcient. In future work we may consider associating

more reﬁned information with patches, exploiting ex-

isting ontologies for modeling additional environment

and ecologic concepts, e.g., ENVO (Buttigieg et al.,

2016) and EcoCore (Buttigieg, 2018).

The Intervention and Operation hierarchies are in-

tended for supporting the planning of improvements

and expansions of the EN. At the current stage of our

work we have modeled them with a few, simple con-

cepts that are sufﬁcient for our purposes. As we ex-

tend our work towards a full-ﬂedged project planning

support system (see section 7), we shall consider ex-

isting works such as (Lazoglou and Angelides, 2016;

SDS Consortium, 2017), which aim at a detailed mo-

deling of the concepts involved in spatial planning and

spatial Decision Support Systems.

Compared to the ontology deﬁned in (Torta et al.,

2017; Torta et al., 2018), the described portion of the

EN Ontology reﬁnes the class ENElement in the types

of EN elements deﬁned in (Bennett and Mulongoy,

2006). Moreover, it establishes a speciﬁc relation be-

tween EN Elements (e.g., Corridor) and the Patches

they are made of. Furthermore, it speciﬁes the land

use characteristics of each patch by associating it with

a LandUseElement.

3.2 Constraints Representation

Constraints in a geographical domain such as ENs

must be able to express restrictions about the objects

of the world modeled by the domain ontology, mix-

ing logic, geometric, and numeric requirements; e.g.,

see (Louwsma et al., 2006). For example, they can

refer to the allowed values of categorical attributes of

areas, to the sum of the sizes of a set of areas, to the

topological relations between pairs of areas, and so

forth.

We have modeled constraints with a Constraint

Ontology, whose classes refer to classes and proper-

ties of the EN Ontology

. We have deﬁned several

kinds of constraints, inspired by the kinds of con-

straints that may appear in a conﬁguration knowledge

base (Soininen et al., 1998; Stumptner et al., 1998;

Felfernig et al., 2002). Figure 4 shows a portion of our

Constraints Ontology, which is structured as a hierar-

chy rooted by class Constraint (in dark gray):

In OWL, referring to classes and properties as values of

other properties is a delicate point; see (W3C, 2017a). We

currently avoid these difﬁculties by exploiting such referen-

ces only in SPARQL queries.

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

140

Figure 4: Diagrammatic representation of a portion of the

constraints hierarchy.

• PartOfConstraint (PartOfCons) refers to a con-

straint that applies to one or more parts of a given

object. Note that a part may be shared by more

than one object; i.e., its semantics is similar to the

aggregation of UML (Object Management Group,

2008).

– SingleAttributeConstraint (SingleAttrCons) in-

volves a single (part-of) attribute of the object.

– MultiAttributeConstraint (MultiAttrCons) in-

volves more than one (part-of) attribute of the

object.

• ConnectionConstraint (ConnectionCons) refers to

a constraint that applies to a relationship between

more than one object.

• PreferenceConstraint (PreferenceCons) repre-

sents soft constraints, that augment regular

constraints with functions to be optimized.

Let us focus on the SingleAttributeConstraints

(SACs). First of all, they refer to a single class (ap-

pliesToClass) and attribute of such class (appliesToAt-

tribute). Note that an attribute of class C is an OWL

property with class C as a possible domain. A spe-

cial kind of SACs applies to attributes that are orde-

red lists of objects/values. In that case, a subset of the

list elements to which the SAC applies can be identi-

ﬁed, through the appliesToItems property. Currently,

this property only allows the speciﬁcation of the par-

ticular class of the items that should be affected by

the constraint. In a future version, we may allow ﬁl-

tering conditions that select such elements based on

their characteristics.

A SAC speciﬁes a condition through the applies-

Condition property; see Figure 5. The main dis-

tinction is between AggregateConditions (Aggrega-

teCond, in gray), that specify restrictions on some

aggregate quantity computed from the elements of

Figure 5: A portion of the condition hierarchy.

a list attribute, and IndividualConditions (Individual-

Cond) that apply to each value of a list attribute, or

to the unique value of a scalar attribute. Individual

conditions are built by composing AtomicConditions

(AtomicCond) into CompositeConditions (Composi-

teCond) with the usual logic connectives and, or, not.

A BinaryCondition has two operands and an ope-

rator (UnaryConditions have an empty secondOpe-

rand). The operands are either constant values (i.e.,

XMLLiterals) or selectors. A selector is modeled as

a list of properties: the list speciﬁes a path from the

element E that is being checked to a value that is re-

achable from E through the chain of properties in the

list.

Note that the the main goal of the representation

is the speciﬁcation of various meta data about con-

straints; e.g., see the distinction between part-of, con-

nection, and preference constraints. The ontology can

be extended and reﬁned as needed for expressing ad-

ditional meta data. This is a key element for the deve-

lopment of reasoners that must automatically retrieve

the suitable constraints to perform constraint solving,

given the characteristics of the input problem.

Example 1. Let us consider the LandscapeCorridor

class in the EN ontology of Figure 2 (see also Fi-

gure 1). A speciﬁcation taken from the guidelines for

the Local EN implementation devised in project (Citt

Metropolitana di Torino, 2014) states that:

Corridors avoid areas with maximum irrever-

sibility and areas with maximum extroversion.

A landscape corridor is therefore made of patches

that must exhibit the speciﬁed characteristics. We can

associate a suitable constraint with class Landscape-

Corridor. The constraint has the following traits:

• it is a SingleListAttributeConstraint, because the

patchList property has a list value;

• it speciﬁes an IndividualCondition that consists of

the conjunction of two BinaryConditions;

Ontological Representation of Constraints for Geographical Reasoning

141

• the ﬁrst BinaryCondition requires a non-

maximum irreversibility; it speciﬁes:

– the ﬁrstOperand as a PropertyList containing

the properties describedBy (from Patch to Lan-

dUseElement) and irreversibility (from LandU-

seElement to the value of the irreversibility cri-

terion);

– the operator as swrlb:greaterThan;

– the secondOperand as the constant value 1.

• the second BinaryCondition is similar to the ﬁrst

one, but it requires non-maximum extroversion.

3.3 Representation of Individual

Information Items

The instances of the domain classes deﬁned in the EN

Ontology, such as the Patches of land that form the

map of a geographic area of interest, are stored in

RDF format (W3C, 2017b) in a triple store that re-

presents the knowledge base used by the system. The

translation from input data-sets (typically available as

ESRI shapeﬁles) to RDF triples is carried out by our

data import functions; see Section 5.

Analogously, the triple store contains the instan-

ces of the constraints classes representing actual con-

straints that apply in the domain, such as the sample

constraint described in Example 1.

4 GEOGRAPHIC REASONING

4.1 The Adjacency Graph Model

We aim to deal with geographic domains, as the Eco-

logical Networks that motivated our work. Thus, it

has been natural to base our reasoning system on a

data model structured as a graph G = (N, E) where:

• the nodes N correspond to areas of a map;

• the arcs E connect nodes whose associated areas

are adjacent in the map.

Figure 6 shows a map and the corresponding ad-

jaciency graph. In the ﬁgure, the graph nodes corre-

sponding to the areas of the map. For example, node

1 corresponds to an area that is adjacent to the areas

corresponding to nodes 2, 3, 5, and 6. Note that, mo-

ving from the map to the graph model, some noise

has been abstracted away; e.g., nodes 5 and 8 are con-

nected although their areas do not actually touch each

other. This abstraction is needed to deal with real-

world, imperfect GIS data, and is a basic task that can

be performed by our system. In particular, when a

Figure 6: A map and its corresponding adjacency graph.

new area A is inserted into the knowledge base of the

system, standard geometric algorithms can be used to:

(i) compute an expansion A

of A consisting of A with

a border of a given thickness, and

(ii) determine the adjacent areas as the ones that in-

tersect with such expanded area A

It is worth pointing out that, while GeoSPARQL

provides a set of functions to compute the Simple Fe-

atures topological relations it deﬁnes, those functions

require that the involved geometries exactly satisfy

the relation, e.g., an area touches another area iff they

share some common points on their borders, but no

internal points. This function is clearly too restrictive

to determine the adjacency of two areas in a meaning-

ful way for our purposes.

As a matter of fact, the graph model is able to ea-

sily accomodate more information than the associa-

tion between nodes and areas:

• Each node n ∈ N can have attributes representing

meta-information about the area A(n) associated

with n. In the EN domain, this may include:

– the values of the evaluation criteria and the LCP

levels of the LandUseElement describing A(n);

– information such as the area size and perimeter

of A(n);

– the identity of the EN Element (e.g., CoreArea)

to which A(n) belongs.

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

142

• Each arc e = (n

, n

) ∈ E can have attributes that

represent meta-information about the relationship

between A(n

) and A(n

). For example:

– the length of the perimeter shared by the two

areas;

– a numeric “cost” of going from A(n

), to A(n

)

determined by the presence of a road/railway

between the two areas.

4.2 Reasoning Tasks

Our model supports reasoning tasks based on three

kinds of inputs:

• the OWL ontologies deﬁning the domain con-

cepts, the constraints hierarchy, and their relati-

onships;

• the RDF data representing the instances of domain

classes of the EN Ontology, and the instances of

the constraints that apply to the speciﬁc domain;

• further requirements provided by the user to spe-

cify the desired reasoning task and its parameters.

Ideally, we would like that the system automatically

extracts all the information needed to perform a re-

quested task from the EN and Constraint Ontologies,

and from the constraint instances, and use it to drive a

generic reasoning engine to compute the answer from

the RDF domain instances. However, such a gene-

rality would be extremely hard, if not impossible to

achieve in practice. Instead, we equip the system with

a pre-deﬁned (but extensible) set of reasoning capabi-

lities that can be reused for several speciﬁc tasks, and

ﬁll the details of speciﬁc reasoning task requests by

exploiting all the input sources mentioned above.

Deﬁnition 1. A Reasoner R (Ω, ∆, ρ) is a function

that takes as inputs an OWL ontology Ω, a RDF graph

∆ and a request ρ and:

1. extracts from ∆ (driven by ontology Ω), a relevant

set of constraints C = C

∪ C

;

2. builds an adjacency graph model G using the con-

straints in C

;

3. performs a reasoning task on G using the con-

straints in C

;

4. returns a result α that answers the request ρ, given

Ω and ∆.

The extraction of constraints is done by issuing

SPARQL queries (W3C, 2017) on the RDF data ∆

built from the vocabulary of ontology Ω; see the next

section for an example. The retrieved constraints are

then returned as data structures that can be directly in-

terpreted by the reasoner in order to perform steps (2)

and (3) above.

Speciﬁc tasks are requested by executing Com-

mands that are translated to one or several invocations

of the reasoners with speciﬁc values of ρ.

In the next subsection, we shall brieﬂy describe

two reasoners that we have implemented in order to

support the automated identiﬁcation, starting from ge-

ographic data about an area, of core areas, and corri-

dors, according to the speciﬁcations given in the EN

Ontology.

4.3 Identifying Core Areas and

Corridors

Reasoner #1. The R

PAT H

reasoner is meant to receive

through the request θ two identiﬁers id

and id

of Pa-

tches or elements that aggregate Patches (i.e., EN ele-

ments), and the name of a property prop that is a list

of Patches or elements that aggregate Patches. It then

computes a path of adjacent elements from element

to id

taking into account the constraints associa-

ted with property prop.

This reasoner can be used to implement the com-

mand BUILD(LandscapeCorridor, id

, id

) which

assigns id

, id

to the from and to attributes of

LandscapeCorridor, and computes the value of the

patchList attribute by invoking R

PAT H

with θ =

(id

, id

, patchList).

1. First of all, the reasoner issues a number of

SPARQL queries to retrieve the SingleListAttribu-

teConstraint C associated with patchList (descri-

bed in Example 1).

2. Then, it builds an adjacency graph G in such a

way that the nodes of G are associated to patches

that satisfy C; i.e., they have non-maximum irre-

versibility and extroversion.

3. Finally, the reasoner applies a simple path-ﬁnding

algorithm based on the well-known A

∗

algorithm

(Dechter and Pearl, 1985) to identify a corridor

between the id

and id

elements, if any.

Reasoner #2. The R

CLUST

reasoner is meant to re-

ceive, through request θ, the identiﬁer id of a Patch,

and the name of a property prop that is a one-to-many

relationship from a class and Patch. Given these in-

puts, the reasoner computes a clustering of patches

that satisfy prop, starting from id, by taking into ac-

count the constraints associated with property prop.

The reasoner can be used to implement the com-

mand BUILD(CoreArea, id), which computes the va-

lue of the madeOf attribute by invoking R

CLUST

with

θ = (id, madeOf ). Note that the madeOf attribute of

CoreArea is associated with a constraint that speciﬁes

the characteristics of the patches P that can be inclu-

ded in a core area. These are the patches with high or

Ontological Representation of Constraints for Geographical Reasoning

143

medium ecological functionality, which is a function

of the naturality and relevance of the LandUseEle-

ment that describes patch P.

5 IMPLEMENTATION

We have implemented the model described in the pre-

vious sections as a Java library consisting of the fol-

lowing modules:

• data-import contains functions supporting the

import/export of shape ﬁles to/from a triple

store (e.g., Parliament (Battle and Kolas, 2012)),

the pre-processing and optimization of the geo-

metries associated with geo-SPARQL Features,

and the transfer of RDF triples between disk

and the in-memory model of the Jena library

(https://jena.apache.org/) used to query the triple

store;

• reasoning contains the functions for the creation

of the adjacency graph data model. Moreover, it

collects all the speciﬁc reasoners provided by the

system (currently, the R

PAT H

and R

CLUST

reaso-

ners described above);

• commands implements the parsing of commands

(currently, the two forms of the BUILD command

described above) and interfaces with the reaso-

ning and data-import functions to execute them;

• shared provides the deﬁnitions and implementati-

ons of elements relevant across the other modules,

such as the geometric feature and triple store ma-

nager, as well as utility functions used by the other

modules.

By exploiting the data-import module, we have po-

pulated the Parliament triple store with 395 patches

deﬁned in the shape ﬁles of a portion of map situated

at the north of the Italian city of Turin. We have then

used the implementation of the R

CLUST

reasoner con-

tained in the reasoning module to generate the Core

Areas as clusters of patches with given characteris-

tics. The reasoner has generated 74 clusters. Finally,

we have used the implementation of the R

PAT H

re-

asoner to generate a number of landscape Corridors

between pairs of Core Areas speciﬁed by us.

6 RELATED WORK

To the best of our knowledge, the only work pur-

suing the automation of a task needed for EN project

planning is the one described in (Torta et al., 2017;

Torta et al., 2018). In that work, the authors deﬁned

a machine readable speciﬁcation of ENs to support

EN validation. They introduced an ad-hoc constraint

satisfaction language for the veriﬁcation of EN con-

straints on a geographical area; e.g., to check whether

a certain area, identiﬁed as a Buffer zone in a pre-

deﬁned EN, complies with the deﬁnition given in the

EN speciﬁcations, or not. While compliance veriﬁca-

tion with respect to an pre-deﬁned EN can be useful to

help the human planners in an EN planning task, our

aim is to take a step further, by addressing the automa-

tic suggestion of suitable aggregations of land patches

into elements of the EN. Moreover, we aim at automa-

tically suggesting modiﬁcations to existing elements

needed by the urban design projects, through suitable

interventions. Even though the current implementa-

tion of our model supports a limited kind of soluti-

ons to tasks only involving two types of elements of

ENs (e.g., the suggestion of modiﬁcations is out of the

scope of the present paper), the proposed approach is

designed to support full ﬂedged implementations of

creation and modiﬁcation tasks.

Another key aspect of our method is the represen-

tation of constraint types as classes of an OWL onto-

logy (the Constraints Ontology). In this way, we em-

ploy a single, well-known standard for the represen-

tation of semantic knowledge, and we avoid the intro-

duction of a new language, requiring ad-hoc tools for

managing constraint information. Even more impor-

tantly, an ontology of constraints supports a detailed

description of the different kinds of constraints (e.g.,

soft and hard, part-of and relational, aggregation and

individual) by qualifying their scope, purpose, relati-

onships, and so forth. This meta-information about

constraints enables the reasoners to retrieve the con-

straints that are relevant for the task at hand, and use

these constraints in the correct places and in the cor-

rect way. This is the reason why we did not adopt

any of the existing ways to represent constraints in

the Semantic Web, including rule languages such

as SWRL (Semantic Web Rule Language) (Horrocks

et al., 2004) and RuleML (Boley et al., 2001), or any

logical/functional languages such as CIF (Constraint

Interchange Format) (Gray et al., 2001). Those lan-

guages allow the deﬁnition of constraints as generic

rules or logic formulas, without characterizing their

properties as needed to automatically retrieve con-

straints and apply them for speciﬁc reasoning tasks.

Regarding the current implementation of the

PAT H

reasoner, it is worth noting that path ﬁnding

has been explored in recommender systems research

to suggest travel paths suiting individual preferences;

e.g., the shortest path between two endpoints, or a

path maximizing pleasure, calm, or other numerical

properties of an area (Quercia et al., 2014). Given

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

144

a graph representing the travel map of a geographi-

cal area, those works choose the path(s) composed of

road segments which, globally, maximize one or more

measures associated to the selected properties. These

approaches solve a speciﬁc task taking into account

a pre-deﬁned set of constraints. Differently, we aim

at supporting multiple reasoning tasks and retrieving

relevant constraints from a semantic knowledge base

(e.g., an RDF store) by exploiting their description as

classes of an OWL ontology.

7 FUTURE WORK

The work presented in this paper builds a solid

ground, both from the modeling and reasoning points

of views, for supporting a user in the project planning

for a geographical domain such as the one of Ecolo-

gical Networks. However, there are several directions

in which our work may be extended to provide a full

ﬂedged decision support system that human users can

effectively use in the deﬁnition of public policies for

land use. For instance, we plan to:

• extend the Constraint Ontology and reﬁne it to

specify more types of constraints, especially soft

constraints for guiding the reasoners to compute

solutions that maximize some preference criteria;

and sophisticated geometric constraints about the

shapes, sizes, etc. of given areas;

• extend our reasoning framework with the ability

of exploiting soft constraints, and with additio-

nal reasoners (e.g., for proposing maintenance and

modiﬁcation interventions on an EN with the goal

of expanding or improving it);

• create a GUI-based application to support the

mixed-initiative interaction between the human

user and the reasoning system to jointly perform

project planning tasks.

8 CONCLUSIONS

We described a semantic framework supporting va-

rious types of constraint-based reasoning tasks on a

geographic domain, from the basic validation of con-

ditions in a geographic area, to the suggestions for de-

ﬁning and modifying EN elements. Different from ot-

her constraint-based geographic reasoners, our model

represents both the geographic domain and its con-

straints in OWL ontologies. This approach has se-

veral advantages: ﬁrst of all, it does not introduce

any ad-hoc language for the management of con-

straints. Second, it fully exploits the knowledge re-

presentation and reasoning interoperability provided

by semantic languages for knowledge sharing and for

data/application interoperability. Third, it opens the

avenue to the classiﬁcation of constraints for their au-

tomated management within reasoners able to adapt

to solve a range of reasoning problems.

As a test-bed for our framework, we considered

the domain of Ecological Networks. Whereas, at the

current stage, we implemented reasoning about ENs

as a stand-alone model, the main motivation and ap-

plication of our work lies in its possible integration

within Participatory Geographical Information Sys-

tems (PGIS, (Sun and Li, 2016)), in order to support

online interaction with stakeholders in inclusive pro-

cesses aimed at collecting feedback and project pro-

posals from stakeholders.

ACKNOWLEDGEMENTS

This work is partially funded by project MIMOSA

(MultIModal Ontology-driven query system for the

heterogeneous data of a SmArtcity, “Progetto di Ate-

neo Torino call2014 L2 157”, 2015-17), and by “Ri-

cerca Locale” of the University of Torino. We thank

Adriano Savoca for his precious contributions to the

initial steps of this work.

REFERENCES

Baglatzi, A., Kokla, M., and Kavouras, M. (2012). Se-

mantifying openstreetmap. In Proc. of Terra Cognita

Workshop on Foundations, Technologies and Applica-

tions of the Geospatial Web - CEUR Workshop Pro-

ceedings, Vol. 901, page paper 4. CEUR.

Battle, R. and Kolas, D. (2012). Enabling the geospatial

Semantic Web with Parliament and GeoSPARQL. Se-

mantic Web, 3(4):355–370.

Benedict, M. and McMahon, E. (2002). Green Infrastruc-

ture: smart conservation for the 21st century. Watch

Clearinghouse Monograph Series. Sprawl.

Bennett, G. and Mulongoy, K. (2006). Review of expe-

rience with ecological networks, corridors and buffer

zones. Technical Series, 23.

Bennett, G. and Wit, P. (2001). The development and appli-

cation of ecological networks: a review of proposals,

plans and programmes. AIDEnvironment.

Boley, H., Tabet, S., and Wagner, G. (2001). Design ratio-

nale of ruleml: A markup language for semantic web

rules. In Proceedings of the First Int. Conf. on Seman-

tic Web Working, pages 381–401. CEUR-WS. org.

Buttigieg, P. L. (2018). Ecology Core Ontology.

https://github.com/EcologicalSemantics/ecocore.

Accessed: 2018-07-28.

Ontological Representation of Constraints for Geographical Reasoning

145

Buttigieg, P. L., Paﬁlis, E., Lewis, S. E., Schildhauer, M. P.,

Walls, R. L., and Mungall, C. J. (2016). The envi-

ronment ontology in 2016: bridging domains with in-

creased scope, semantic density, and interoperation.

Journal of biomedical semantics, 7(1):57.

Citt

a Metropolitana di Torino (2014). Mi-

sura 323 del PSR 2007-2013 (in Italian).

http://www.cittametropolitana.torino.it/cms/territorio-

urbanistica/misura-323/misura-323-sperimentale.

Codescu, M., Horsinka, G., Kutz, O., Mossakowski, T., and

Rau, R. (2011). OSMonto – an ontology of openstreet-

map tags. In Proc. State of the map Europe (SOTM-

EU) 2011, pages 55–64, Vienna, Austria.

Council of Europe (2000). General guidelines for the deve-

lopment of the pan-european ecological network. Na-

ture and environment, 107.

Dechter, R. and Pearl, J. (1985). Generalized best-ﬁrst se-

arch strategies and the optimality of a*. Journal of the

ACM, 32(3):505536.

Fath, B., Sharler, U., Ulanowicz, R., and Hannon, B. (2007).

Ecological network analysis: network construction.

Trends in Ecology & Evolution, 208:49–55.

Felfernig, A., Friedrich, G., Jannach, D., and Zanker, M.

(2002). Conﬁguration knowledge representation using

uml/ocl. In International Conference on the Uniﬁed

Modeling Language, pages 49–62. Springer.

Fonseca, F., Egenhofer, M., Agouris, P., and C

amara, G.

(2002a). Using ontologies for geographic information

systems. Transactions in GIS, 3:231–257.

Fonseca, F., Egenhofer, M., C.A. Davis, and C

amara, G.

(2002b). Semantic granularity in ontology-driven ge-

ographic information systems. Annals of Mathematics

and Artiﬁcial Intelligence, 36(1-2):121–151.

GeoNames.org (2018). Geonames mappings ontology.

http://www.geonames.org/ontology/mappings v3.01.rdf.

Gobluski, A., Westlund, E., Vandermeer, J., and Pascual,

M. (2016). Ecological networks over the edge: Hyper-

graph trait-mediated indirect interaction (TMII) struc-

ture. Trends in Ecology & Evolution, 31(5):344354.

Gray, P., Hui, K., and Preece, A. (2001). An expressive

constraint language for semantic web applications. In

E-Business and the Intelligent Web: Papers from the

IJCAI-01 Workshop, pages 46–53.

Hall, M. and Mandl, P. (2006). Spatially extended ontolo-

gies for a semantic model of harmonised landuse and

landcover information.

Horrocks, I., Patel-Schneider, P., Boley, H., Tabet, S., Gro-

sof, B., and Dean, M. (2004). SWRL: A semantic web

rule language combining OWL and RuleML. W3C

Member submission, 21.

Janowicz, K., Scheider, S., Pehle, T., and Ha, G. (2012a).

Geospatial semantics and linked spatiotemporal data

past, present, and future. Semantic Web - On linked

spatiotemporal data and geo-ontologies, 3(4):321–

332.

Janowicz, K., Scheider, S., Pehle, T., and Ha, G. (2012b).

Linkedgeodata: A core for a web of spatial open data.

Semantic Web Interoperability, Usability, Applicabi-

lity, 3(4):333–354.

Jongman, R. (1995). Nature conservation planning in eu-

rope: developing ecological networks. Landscape and

urban planning, 32(3):169–183.

Lazoglou, M. and Angelides, D. C. (2016). Development

of an ontology for modeling spatial planning systems.

Current Urban Studies, 4(02):241.

Louwsma, J., Zlatanova, S., van Lammeren, R., and van

Oosterom, P. (2006). Specifying and implementing

constraints in GIS - with examples from a geo-virtual

reality system. GeoInformatica, 10(4):531–550.

Lurgi, M. and Robertson, D. (2011). Automated experimen-

tation in ecological networks. Automated Experimen-

tation, 3(1).

Montenegro, N., Gomes, J., Urbano, P., and Duarte, J.

(2011). An owl2 land use ontology: Lbcs. In Interna-

tional Conference on Computational Science and Its

Applications, pages 185–198. Springer.

Object Management Group, O. (2008). Uniﬁed Modeling

Language (UML. http://www.uml.org.

OGC (2012). Geosparql vocabulary. http://schemas.

opengis.net/geosparql/1.0/geosparql vocab all.rdf.

Open Geospatial Consortium et al. (2011). OpenGIS Im-

plementation Standard for Geographic information-

Simple feature access-Part 1: Common architecture.

OpenStreetMap Contributors (2017). Openstreetmap.

https://www.openstreetmap.org.

Palacio, D., Derungs, C., and Purves, R. (2015). Develop-

ment and evaluation of a geographic information re-

trieval system using ﬁne grained toponyms. Journal

of Spatial Information Science JoSIS, 11:1–29.

Pilosof, S., Porter, M., Pascual, M., and Keﬁ, S. (2017).

The mulutilayer nature of ecological networks. Nature

ecology&evolution, 1:article 101.

Provincia di Torino (2014). Linee guida per le reti

ecologiche (in italian). http://www.provincia.

torino.gov.it/territorio/ﬁle-storage/download/pdf/

pian territoriale/rete ecologica/lgsv lgre.pdf.

Quercia, D., Schifanella, R., and Aiello, L. (2014). The

shortest path to happiness: Recommending beautiful,

quiet, and happy routes in the city. In Proc. of the 25th

ACM Conference on Hypertext and Social Media, HT

’14, pages 116–125, New York, NY, USA. ACM.

SDS Consortium (2017). Spatial decision support ontology.

http://sdsportal.sdsconsortium.org/ontology/.

Soininen, T., Tiihonen, J., M

annist

o, T., and Sulonen, R.

(1998). Towards a general ontology of conﬁguration.

Ai Edam, 12(4):357–372.

Stumptner, M., Friedrich, G. E., and Haselb

ock, A. (1998).

Generative constraint-based conﬁguration of large

technical systems. AI EDAM, 12(4):307–320.

Sun, Y. and Li, S. (2016). Real-time collaborative GIS: a

technological review. ISPRS Journal of Photogram-

metry and remote sensing, 115:143–152.

Torta, G., Ardissono, L., Corona, M., La Riccia, L., Sa-

voca, A., and Voghera, A. (2018). GeCoLan: a con-

straint language for reasoning about ecological net-

works in the semantic web. In Knowledge Disco-

very, Knowledge Engineering and Knowledge Mana-

gement, IC3K 2017, Revised Selected Papers, page

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

146

to appear. Springer-Verlag, Berlin Heildelberg New

York.

Torta, G., Ardissono, L., Savoca, A., Voghera, A., and

Riccia, L. L. (2017). Representing ecological net-

work speciﬁcations with semantic web techniques. In

Proc. of 9th Int. Joint Conf. on Knowledge Discovery,

Knowledge Engineering and Knowledge Management

(KEOD 2017), pages 86–97, Funchal, Madeira, Portu-

gal. SCITEPRESS.

Ulanowicz, R. (2004). Quantitative methods for ecological

network analysis. Computational biology and chemi-

stry, 28:321339.

Van Hage, W. R., Malais

e, V., Segers, R., Hollink, L., and

Schreiber, G. (2011). Design and use of the simple

event model (sem). Web Semantics: Science, Services

and Agents on the World Wide Web, 9(2):128–136.

Voghera, A. and La Riccia, L. (2019). Ecological networks

in urban planning: Between theoretical approaches

and operational measures. In Calabr

o, F., Della Spina,

L., and Bevilacqua, C., editors, New Metropolitan Per-

spectives, pages 672–680, Cham. Springer Internatio-

nal Publishing.

W3C (2017a). Representing classes as property values

on the semantic web. https://www.w3.org/TR/2005/

NOTE-swbp-classes-as-values-20050405/.

W3C (2017b). Resource description framework (RDF).

https:// www.w3.org/RDF/.

W3C (2017). SPARQL query language for RDF. https://

www.w3.org/TR/rdf-sparql-query/.

W3C (2017). Web ontology language (OWL). https://

www.w3.org/OWL/.

Ontological Representation of Constraints for Geographical Reasoning

147