A Multi-level Ontological Approach for Change Monitoring in Remotely

Sensed Imagery

Fethi Ghazouani

, Wassim Messaoudi

and Imed Riadh Farah

1,2

RIADI Laboratory National School of Computer Science University of Manouba, Tunisia

I.T.I Departement, Telecom Bretagne Technop

ole Brest Iroise, CS 83818, Brest, France

Keywords:

Spatio-temporal Object, Dynamics Object, Change Detection, Domain Ontology, Upper Ontology, Multi-

level.

Abstract:

Land-use/cover change, climate change, sea level evolution are examples of application that are associated

with change detection. Actually, we use satellite image time series to monitor the change where entities are

often dynamic along time. Moreover, knowledge associated to these spatio-temporal objects can evolve when

changes occur. Thus, for modeling this kind of knowledge it is necessary to deal with four aspects: spectral,

spatial, temporal and semantic. Such approach can be modeled by ontologies in many levels. Thereby, a

shared ontology can be an ontology or a combination of some ontologies based on some mechanisms of

linking. Such link process should maintain consistency between represented knowledge. In this paper, we

propose a multi-level ontological approach for monitoring dynamics in remote sensing images. The proposed

methodology aims to link our domain ontology to an upper level ontology thus enabling to represent existing

change processes.

1 INTRODUCTION

Change detection is one of the main applications of

remote sensing community. Singh (Singh, 1989) de-

ﬁned change detection as the process of identifying

differences in the state of an object or phenomenon by

observing it at different times. The remotely sensed

data become a major source for change detection and

monitoring studies because of its high temporal fre-

quency, digital format suitable for computation, syn-

optic view, and wider selection of spatial and spectral

resolutions (Hussain et al., 2013). Geographic objects

in these data images are often dynamic. Indeed, ob-

jects with spatial representation might grow, shrink

(the case of the urban space), change their shape (for

example when a city changes its borders), divide (in

the case when a forest is divided into urban zone and

forest), disappear (a lake can disappear) or merge into

a new object (sub-parcels are merged into one parcel)

in time. Knowledge associated to spatio-temporal ob-

ject can evolve when changes occur on thematic at-

tributes of objects. The semantic dimension of an en-

tity aims to describe the knowledge associated with

the entity. Adding semantic capabilities to GIS tools

is one of solution that allows handling the semantics

of the spatial-temporal objects. However, this solu-

tion does not offer the capacity to perform inference

or reasoning on information from spatio-temporal dy-

namic phenomena (Harbelot et al., 2013).

Thus, a better framework must be able to repre-

sent the description of the knowledge in an enhanced

way that can be used to perform reasoning on dy-

namic phenomena. Such approach is offered by on-

tologies. Ontology allows a formal representation of

knowledge as well as on the reasoning on this knowl-

edge. This paper is organized as follows. In Sec-

tion 2, we start by introduce the basic principle of

ontology as a knowledge representation technology

and presents the different types of ontologies, then we

present a state of the art of different existed ontology-

based approaches for modeling dynamics of objects

and ﬁnally, we describe the dynamics in remote sens-

ing. In Section 3, illustrate the proposed approach for

modeling change in remote sensing images. Finally,

we present our conclusions in Section 4.

2 MODELING DYNAMICS

2.1 Ontology

Recently, ontology is considered as one of the better

Ghazouani, F., Messaoudi, W. and Farah, I..

A Multi-level Ontological Approach for Change Monitoring in Remotely Sensed Imagery.

In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 2: KEOD, pages 435-440

ISBN: 978-989-758-158-8

435

techniques used for the interpretation of images. Gru-

ber (Gruber, 1995) deﬁnes ontology as a formal spec-

iﬁcation of a shared conceptualization. Ontologies

specify a set of concepts, instances, relationships, and

axioms that are relevant for modeling a domain of

study (Gruber, 1995) and permit the inference of im-

plicit knowledge. Nevertheless, ontologies can be

very different both in terms of their ”top-level” as at

the level of the treatment of their basic components

such things, process, relationships, etc. Thus, ontolo-

gies can be classiﬁed according to two dimensions:

level of detail and degree of dependence relative to a

particular task or a point of view. Precisely, (Guarino,

1998) classiﬁes ontologies according to their general-

ity levels. At the top level, upper ontologies describe

general concepts or sense knowledge such as space,

time, materiel, objects, events, actions, etc., which are

independent to a deﬁned problem or to a particular

application domain. These ontologies provide gen-

eral concepts to which all terms of existing ontologies

must be linked. Domain ontologies are specialized for

a certain type of artifact. They describe the vocabu-

lary related to a generic domain (such as medicine or

automobiles) by specializing the concepts presented

in high level ontologies. Task ontologies describe vo-

cabulary related to a task or a generic activity (such

as diagnosis or sale). These ontologies provide a sys-

tematic lexicon of terms used to solve the problems

associated with particular tasks (dependent or not to

the domain).

2.2 Modeling Dynamics with Upper

Ontologies

Spatial-temporal representations are offered, by so-

called upper ontologies (foundation ontologies), such

as DOLCE (Masolo et al., 2003), BFO (Grenon and

Smith, 2004), GFO (Herre, 2010) and others. Founda-

tion ontologies provide a meta-language (Mizoguchi

et al., 1995) to ontological approach which allows to

model spatio-temporal phenomena.

In these ontologies, there exists a fundamental

distinction between static entities (continuants or en-

durants) and dynamic entities (occurrents or perdu-

rants). Endurants are objects which persist over time.

They include physical objects, for example: tree, lake,

and river. Perdurants are objects which are ”happen-

ing” at the time. They include events or processes,

but some systems, like BFO, extend the list by adding

temporal and spatio-temporal regions.

Probst (Probst, 2006) have presented an ontologi-

cal analysis of observations and measurements for as-

sessing semantic interoperability between geospatial

information sources. This approach consists to align

the observations and measurements domain ontology

to the foundation ontology DOLCE. The alignment

is performed by the interpretations of the central ele-

ments of the observations and measurements concep-

tual model in the DOLCE context and establishes ex-

plicit relations between categories of real world enti-

ties and classes of information objects. For example,

a phenomenon is described by an observed property

in the the observations and measurements speciﬁca-

tion. In DOLCE, only qualities are entities that can be

observed. In this case, the category Observed Prop-

erty in the observations and measurements model sub-

sumes the category Quality in DOLCE (Figure 1).

Figure 1: Aligning the domain concept observation to the

concept quality of DOLCE (Probst, 2006).

In the medical domain, Camara (Camara et al.,

2012) has proposed an ontological approach for mon-

itoring and preventing the propagation of infectious

diseases. The conceptual framework adopted for

building the propagation ontology of infectious dis-

eases is structured in three layers: (i) foundation layer

which contains the upper ontology BFO, (ii) the core

layer that constitutes the IDO-core ontology where

IDO (Infectious Domain ontology) is a domain on-

tology, and (iii) the speciﬁc layer contains the sub-

domains ontologies. To align those three layers, au-

thors start by a categorization of the domain entities

as continuants or occurrents. Then, based on the IDO-

core ontology, they have linked categorized concepts

to their equivalent or parent concepts on the IDO-core

which are in turns connected to BFO. In the same do-

main, Weichert (Weichert et al., 2013) has introduced

a temporal domain ontology for biomedical simula-

tions. Aligning the domain ontology to BFO allows

to monitor the dynamics of blood ﬂow simulations.

The correlating categories from both ontologies (BFO

and domain ontology) are identiﬁed and connected by

newly inserted is a relations.

To facilitate mutual understanding between re-

searchers, managers and local communities, Duce

(Duce, 2009) has proposed an informal ontology for

reef islands. Such ontology allows to aid the appli-

cation of information technologies to the forecasting

and monitoring of climate-change-related to impact in

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

436

reef islands environments. Indeed, the author has de-

veloped a small, prototypical reef island domain on-

tology, based on informal, natural language relations,

for 20 fundamental terms within the domain. Then,

in order to create a coherent, systematic and complete

ontology, the domain ontology have been aligned to

DOLCE as a top level ontology. To align domain on-

tology to upper ontology, the author has selected a

subset of particulars from the reef island domain in

accordance with classes categories of the upper-level

DOLCE ontology.

In a similar way, Devaraju (Devaraju and Kuhn,

2010) propose a process-based ontology for repre-

senting dynamic geospatial phenomena. The pro-

posed approach aims to improve water planning

and management through continuous monitoring and

forecasting of river ﬂow. This approach allows to use

DOLCE top-level ontology to ease and guide the rep-

resentation of foundational entities needed to repre-

sent dynamic phenomena. Authors have aligned the

domain concepts describing precipitation and evap-

otranspiration (evaporation and transpiration) pro-

cesses to the general categories deﬁned in DOLCE.

Thus, the entities will be identiﬁed and assigned to

perdurant, endurant and quality notions in the top-

level ontology.

An interesting work that shows the effectiveness

of the upper ontologies, in the remote sensing domain,

has been performed by Kauppinen and de Espindola

(Kauppinen and de Espindola, 2011). In their work,

they are interested in giving an ontological foundation

to essential land change trajectories, and to modeling

them with formal semantics. To achieve this they pro-

pose the Process-oriented Land Use and Tenure On-

tology (PLUTO), built as an alignment to the top-level

ontology DOLCE, for semantically integrating sev-

eral data sets related to deforestation and land change

trajectory in the Brazilian Amazon.

2.3 Dynamics in Remotely Sensed

Images

Geographic objects such as lakes, rivers, and storm

fronts have very spatial dynamic properties. It is pos-

sible that an object changes their attributes and spatial

representation along time. This observation is partic-

ularly true on the case of the remote sensing because

major applications are associated with change detec-

tion such as land-use or land-cover change. Land-

cover refers to the observed biotic and abiotic assem-

blage of the earth’s surface and immediate subsur-

face (Meyer and Turner, 1992). Examples of ma-

jor land-cover types are forests, shrublands, grass-

lands, croplands, barren lands, ice and snow, urban

areas, and water bodies. Such information is ob-

tained from ground surveys or through remote sens-

ing. Land-cover change can be characterized as land-

cover conversion and modiﬁcation processes. Land-

cover conversion is a change from one land-cover

category to another, and modiﬁcation is a change in

condition within a land cover category (Meyer and

Turner, 1994), i.e., the more subtle changes that af-

fect the character of the land cover without changing

its global classiﬁcation. An example of the former is

change from cropland to urban land, and an example

of the latter is degradation of forests. Forest degrada-

tion may be due to change in phenology, biomass, for-

est density, canopy closure, insect infestation, ﬂood-

ing, and storm damage.

Remotely sensed images provide measurement

and observation that can be used for monitoring dy-

namics change. Indeed, indices derived from satel-

lite data are widely used for land-cover change stud-

ies. NDVI (Normalized Difference Vegetation index)

values strongly correlate with green vegetation, and

changes in NDVI indicate changes in biological activ-

ities (Verbesselt et al., 2010). NDVI decreases signif-

icantly after green biomass is removed, so it is widely

used for mapping and monitoring ﬁre disturbance,

forest clear-cut activity, urbanization, and other land-

cover changes (Verbesselt et al., 2010). The EVI (en-

hanced vegetation index) has been used for postﬁre

forest regeneration and phenological analysis during

change detection (Liang et al., 2011). Normalized

Difference Fraction Index (NDFI) (Jr. et al., 2005) is

a new spectral index for enhanced detection of forest

canopy damage caused by selective logging and forest

ﬁres. High NDFI values indicate the presence of in-

tact forest, whereas a degraded forest is obtained by a

decreased value of this indice. Other spectral indices

are also used for detecting changes in remotley sensed

images, for more we can refer to (Chandra, 2012).

Thus, it is necessary to model this information in

order to monitor and detect the type of change pre-

sented in satellite images. In following section, we

present the adopted architecture for modeling dynam-

ics objects of remotely sensed images.

3 THE MULTI-LEVEL

ONTOLOGICAL APPROACH

As we have mentioned above, change detection is the

process of identifying differences in the state of a fea-

ture or phenomenon by observing it at different times.

In remote sensing it is useful in land use/land cover

change analysis such as monitoring deforestation or

vegetation phenology.

A Multi-level Ontological Approach for Change Monitoring in Remotely Sensed Imagery

437

Thus, we propose to represent and model the

change process in remote sensing images while bas-

ing on the cited works that adopt multi-level onto-

logical architecture for modeling the dynamics. The

conceptual framework of our suitable multi-level on-

tological architecture, as we illustrate in Figure 2, is

structured on three layers: (1) Fundamental-layer, (2)

Core-layer and (3) Domain-layer. These layers are

described on following.

Figure 2: Multi-level ontological approach.

3.1 The Fundamental Layer

The fundamental layer contains the upper ontology

that we have used in our model. We have chosen

the fundamental ontology BFO (Basic Fundamental

Ontology) as a suitable upper ontology that provides

concepts and relations that may be reused for the con-

struction or for the enrichment of domain ontology.

Consequently, we have reused these abstract concepts

and relations for modeling the ontology of change

processes (core-layer). The choice of the BFO is

based on three criteria: (i) BFO extends the list of

occurrents (event or process) by temporal and spatio-

temporal regions, (ii) the coherence of the categoriza-

tion of the concepts of process, event, state and object

towards their semantic in the domain of remote sens-

ing, and (iii) the consistency of the reuse of relations

between these concepts to cover the speciﬁc relations

in our domain.

3.2 The Core Layer

The core ontology models general concepts and re-

lations related to change processes in remote sensing

images. The deﬁnitions and the classiﬁcation of these

concepts are based on the categorical classes deﬁned

in the BFO ontology. In the following, we present the

different steps for the construction of the core ontol-

ogy of change processes.

• Identify and Categorize Changes Processes:

This step consists for understanding and detecting

modiﬁcations process in addition to conversions. In-

deed, as we have mentioned in Section 3.2, there are

generally distinctions between land cover conversion

process, namely the complete replacement of a type

of cover by another and the land cover modiﬁcation

process i.e., the more subtle changes that affect the

character of the land cover without changing its global

classiﬁcation. Deforestation, urbanization and deser-

tiﬁcation are examples of land cover/land use change

processes.

Deforestation is the conversion of forested areas

to non-forest land use such as arable land, urban use,

logged area or wasteland. According to FAO (Food

and Agriculture Organization), deforestation is the

conversion of forest to another land use or the long-

term reduction of tree canopy cover below the 10%

threshold. Desertiﬁcation is a speciﬁc expression of

land degradation processes. The degradation is a pro-

cess leading to a ”temporary or permanent deterio-

ration in the density or structure of vegetation cover

or its species composition” (Grainger, 1993). In this

case, changes affect soil characteristics, then we con-

sider desertiﬁcation as a modiﬁcation process.

So, in this step we have categorized the changes

as conversion and modiﬁcation processes and, then

each example of change is classiﬁed to one of these

categories. In Figure 3, we illustrate an example of a

change process classiﬁcation.

Figure 3: Process categorization and classiﬁcation.

• Categorize Domain Entities:

This step aims to identify and classify domain entities

that participate in such change process, i.e., to know

what are the main ecological and socio-economic

variables which drive the land-cover change process.

In other words, what are the basic features required

for modeling dynamic phenomena and how they are

classiﬁed? Indeed, entities can be biophysical ob-

jects, features such as biomass, state of vegetation,

soil moisture, ﬁre, etc. These entities will be classi-

ﬁed into two categories of concepts: Continuants (En-

durants) and Occurrents (Perdurants). Continuants

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

438

correspond to entities without temporal part. They

include physical objects that persist in time such as

tree, lake, river, etc. The concepts of occurrents cor-

respond to entities taking place in the time during dif-

ferent phases. They are objects that occur in time and

they include events (such as ﬁre) and processes (cut-

ting down trees) that involve the continuants.

• Identify Categorical Relations:

This step consists to know what are the categorization

relations that hold between concepts (continuants and

occurents). Examples of these relations are :

Participation: A continuant participates in a oc-

current (Grenon and Smith, 2004). A ﬁre participates

in a deforestation.

Parthood: It reﬂects the notion of relationship

”part-of” and it applies to both continuants and oc-

currents. For example, an event is constituted of pro-

cesses.

The Hierarchy: It models the inheritance relation-

ships. As example, a desertiﬁcation is a modiﬁcation

process.

The Causality: It expresses, for example, the fact

that an event ”causes” another event. Less soil mois-

ture causes a less soil canopy.

3.3 The Domain Layer

The domain layer contains the domain ontology of

remote sensing images (Messaoudi et al., 2014) that

represents the domain concepts and their relations.

This ontology allows interpretation and representa-

tion of objects existing in the scene of satellite im-

age. The remote sensing provides observation and

measurement of indicators associated to these objects.

Indicators such as NDVI, NDFI, and derived vari-

ables (entities) of surface composition (soil moisture,

biomass, etc.) are relevant information for detecting

participated concepts to each change process. Figure

4 presents an example of the association of the do-

main concept forest to each of both change processes

(deforestation or degradation) in function of the mea-

sure of its indicator NDFI. In this example, the prop-

erty NDFI of the concept forest indicates that there is

a deforestation process with an NDFI value val1, al-

though it is a degradation process with an NDFI value

val2.

3.4 Alignment of the Three Ontological

Layers

As we have presented in Figure 2, the proposed model

is structured in three layers. The fundamental-layer

contains the upper ontology BFO, the core-layer de-

scribes the core ontology of change processes and the

Figure 4: Association of participants entities.

domain-layer that contains the domain ontology of re-

mote sensing images. It is necessary to link those lay-

ers for representing changes processes in satellite im-

ages. Thus, to align these three layers, we have fol-

lowed the same principle of alignment proposed in the

previously cited works. Indeed, the alignment mecha-

nism consists to identify relations R that can hold be-

tween different concepts C in the three levels. Given

the following formalization of our model, we adopt

the procedure described below for linking the three

levels.

Model =



, O

, C, R



; where:

• O

, O

and O

represent respectively the funda-

mental, the core and the domain ontology.

• C =



, C



; where C

, C

represent re-

spectively a set of fundamental, core and domain

concepts.

• R =

{

subsumption, inherence, parthood,

participation

}

(the set of relations) (cf. subsec-

tion 3.2)

procedure alignment(O

, O

)

// alignment of O

and O

for each change process in O

{

participants (C

, C

)

// the set of participants concepts (do-

main concepts) that participates on the deﬁnition of

a process (core concept) C

in the C

hold relation (C

, C

, R ) // Iden-

tify type of relationship between concepts (causes,

equivalent, participate, etc)

// create the relation link

}

// alignment of O

and O

for each C

in O

{

hold relation (C

, C

, R )

// create the relation link

}

A Multi-level Ontological Approach for Change Monitoring in Remotely Sensed Imagery

439

An example that illustrates the mechanism of

alignment has been presented in Figure 2 and Figure

3. Indeed, the Figure 3 shows the alignment of the

domain layer to core layer. In this ﬁgure, the domain

concept forest participates in the change process de-

forestation (respectively degradation). Then, the con-

cept forest is linked to the concept deforestation with

the relationship participates. Figure 2 shows how the

core layer is linked to the fundamental layer. Indeed,

concepts like conversion and modiﬁcation have been

classiﬁed as two types of processes based on the cate-

gorical classes deﬁned in the BFO ontology (cf. sub-

section 3.2). Thus, the basic concept process in BFO

subsumes respectively the concepts conversion (a pro-

cess) and modiﬁcation. This implies that the relation

that holds between these concepts is a subsumption

(is a) relationship.

4 CONCLUSIONS

Remote sensing is a unique monitoring tool that pro-

vides access to dynamic environments. The essen-

tial, however, is the understanding of processes such

as deforestation, desertiﬁcation, urbanization, etc. A

semantic description of each process enables to iden-

tify concepts, features and relations that hold between

them implying as well that process. In this paper,

we have used a multi-level model based on ontolo-

gies for representing this knowledge enabling thus to

reason on change processes in remotely sensed im-

agery. This model is based on a domain ontology of

remote sensing, core ontology and the upper ontology

BFO. The core ontology represents classiﬁcation and

categorization of different processes of changes.The

domain ontology provides observations and measure-

ments that allow reasoning on such change process

represented in the core ontology. Finally, the ontology

BFO provides concepts and relations that are used to

construct and enrich the core ontology.

REFERENCES

Camara, G., Despr

es, S., Djedidi, R., and Lo, M. (2012).

Vers une ontologie des processus de propagation des

maladies infectieuses. In IC, pages 99–114, Paris,

France.

Chandra, P. (2012). Remote Sensing of Land Use and Land

Cover: Principles and Applications. CRC Press.

Devaraju, A. and Kuhn, W. (2010). A process-centric on-

tological approach for integrating geo-sensor data. In

FOIS, volume 209, pages 199–212, Toronto, Canada.

Duce, S. (2009). Towards an ontology for reef islands. In

GeoSpatial Semantics, volume 5892, pages 175–187,

Mexico City, Mexico.

Grainger, A. (1993). Controlling tropical deforestation.

Earthscan Publications Ltd, page 310.

Grenon, P. and Smith, B. (2004). Snap and span: Towards

dynamic spatial ontology. Spat.Cog.Comp., 4:69–104.

Gruber, T. R. (1995). What is an ontology. International

Journal of human Computer Studies, 43:907–928.

Guarino, N. (1998). Formal ontology and information sys-

tems. In FOIS’98, pages 3–15, Trento, Italy.

Harbelot, B., Arenas, H., and Cruz, C. (2013). A semantic

model to query spatial- temporal data. In the 6th In-

ternational Workshop on Information Fusion and Ge-

ographic Information Systems, Petersburg, Russia.

Herre, H. (2010). General formal ontology (gfo) a foun-

dational ontology for conceptual modelling. Media,

2:1–50.

Hussain, M., Chen, D., Cheng, A., Wei, H., and Stanley, D.

(2013). Change detection from remotely sensed im-

ages: From pixel-based to object-based approaches.

ISPRS, 80:91–106.

Jr., C. M. S., Robertsb, D. A., and Cochrane, M. A. (2005).

Combining spectral and spatial information to map

canopy damage from selective logging and forest ﬁres.

Remote Sensing of Environment, 98:329–343.

Kauppinen, T. and de Espindola, G. M. (2011). Ontology-

based modeling of land change trajectories in the

brazilian amazon. In Geoinformatik.

Liang, L., Schwartz, M., and Fei, S. (2011). Validating

satellite phenology through intensive ground observa-

tion and landscape scaling in a mixed seasonal forest.

Remote Sensing of Environment, 115:143–157.

Masolo, C., Borgo, S., Gangemi, A., Guarino, N., Oltra-

mari, A., and Schneider, L. (2003). The wonderweb

library of foundational ontologies. Preliminary report,

WonderWeb Deliverable D17.

Messaoudi, W., Farah, I. R., and Solaiman, B. (2014).

A new ontology for semantic annotation of remotely

sensed images. In ATSIP, Sousse, Tunisia. IEEE.

Meyer, W. and Turner, B. (1992). Human-population

growth and global land-use cover change. Annual Re-

view of Ecology and Systematics, 23:39–61.

Meyer, W. and Turner, B. (1994). Changes in land use

and land cover: A global perspective. OIES Global

Change Institute, pages 33–34.

Mizoguchi, R., Vanwelkenhuysen, J., and Ikeda, M. (1995).

Task Ontology for Reuse of Problem Solving Knowl-

edge, pages 46–59. Amsterdam.

Probst, F. (2006). An ontological analysis of observations

and measurements. M

unster, Germany. GIScience.

Singh, A. (1989). Digital change detection techniques using

remotely-sensed data. IJRS, 10:989–1003.

Verbesselt, J., Hyndman, R., Zeileis, A., and Culvenor, D.

(2010). Phenological change detection while account-

ing for abrupt and gradual trends in satellite image

time series. RSE, 114:2970–2980.

Weichert, F., Mertens, C., Walczak, L., Kern-Isberner, G.,

and Wagner, M. (2013). A novel approach for con-

necting temporal-ontologies with blood ﬂow simula-

tions. JBI, 46:470–479.

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

440