MULTI-AGENT BASED MODELING OF THE TUNISIAN

PASTORAL DYNAMIC

Multi-level Organization

Islem Henane

1,3

, Sameh Hadouaj

2,3

and Khaled Ghedira

1,3

Higher Institute of Management, Tunis, Tunisia

Faculty of Economic Sciences and Management, Nabeul, Tunisia

Laboratory of Strategy Optimisation of Information and Knowledge (SOIE), Tunis, Tunisia

Keywords: Multi-agent Based Modeling, Complex System, Multi-level Organisation, Emergence, Multi-agent

Platforms.

Abstract: Pastoral systems in arid and semi arid areas are characterized by a continued deterioration. This degradation

is the result of the mismanagement of resources in response to natural, economic and social mutations.

These systems are considered as complex systems, given the large number of stakeholders in interaction and

levels of granularity. To address this situation, analytical and systemic approaches are no longer adequate.

In this paper, we propose a multi-agent based model of Tunisian pastoral dynamics taking into account the

interaction dynamics of the different stakeholders and the different levels of granularity. The completion of

this work is within the scope of the development of the Intelligent Decision Support System PASDES

(Pastoral Strategies Definition System). PASDES aims to support pastoral strategic decision making in short

and long terms.

1 INTRODUCTION

Tunisian rangelands spread over almost one third of

the country's area. Almost half of these rangelands

are collectives. These figures reflect the importance

of the pastoral activity. However, the Tunisian

pastoral resources are in a continued deterioration.

This deterioration is caused by the inappropriate

strategies undertaken to address climate (high

temperature, unstable precipitation) economical

(increase in grain prices, global crisis) and social

changes (privatization, competition over resources).

The sustainability of the system is threatened.

The pastoral system is a complex system

characterised by a large number of interacting

entities (plants, animals, shepherds, economic actors,

state, etc) and different levels of granularity (

vegetation dynamics, animal dynamics, shepherds

interactions and negotiations, etc). Thus, the study of

such a system is conducted in the context of solving

complex problems. In literature, three approaches

are mainly used to study such a system. First, the

analytical approach was adopted by economists. It

focuses on the elementary study of the system

components. For example, the study can be based on

the maximizing of an objective function on a

particular element of the system (Lalba, Zoundi and

Tiendrebeogo, 2005) (Dutilly-Diane, 2006). This

approach deals with one variable at a time which

prevents the study of the overall system dynamics.

Second, the systemic approach was adopted by

ecologists. It is based on the study of the overall

system behavior (Costanza, Wainger, Folke and

Mäler, 1993) without taking into account the micro-

dynamics. In this context, the emergence of global

properties is evoked (Bergandi , 2000) (Oprisan and

Oprisan, 2006). Third, the constructivist approach

brings together the contributions of the analytical

(micro-level study) approach as well as the systemic

approach (macro- level study). It focuses on the

study of micro-order interactions of different entities

as well as the global system behavior. In this

approach, we can distinguish the use of cellular

automata (Soares-Filho, Cerqueira and Pennachin,

2002) and multi-agent systems. When using cellular

automata, the environment is represented by a set of

cells. The behavior of individuals is defined by a

finite set of states, transition rules and

neighbourhood relationship. However, cellular

254

Henane I., Hadouaj S. and Ghedira K..

MULTI-AGENT BASED MODELING OF THE TUNISIAN PASTORAL DYNAMIC - Multi-level Organization.

DOI: 10.5220/0003832802540258

In Proceedings of the 4th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2012), pages 254-258

ISBN: 978-989-8425-96-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

automata are unable to take into account the

complexity and heterogeneity of stakeholders’

behaviors. On the other side, multi-agent systems are

characterised by their ability of modeling

interactions between autonomous agents and with

their environment. In this work, we opted for the use

of multi-agent systems. In fact, using multi-agent

systems in a complex system solving context takes

benefits of (1) ease of modeling of distributed

systems (2) possibility of modeling of the cognitive

decision processes (3) explanatory ability of

dynamics and micro and macro variables (self-

organization, emergence). The completion of this

work is within the scope of the development of the

Intelligent Decision Support System PASDES

(PAstoral Strategies Definition System). PASDES

aims to support pastoral strategies decision making

in short and long terms. In this paper, we propose a

multi-level modeling of the Tunisian pastoral system

taking into account the interaction dynamics of

different stakeholders and climatic, economic and

social variables. Such multi-level organization

allows design and modeling of each level in its own.

It allows understanding more the system dynamics.

When modeling all levels of granularity is achieved,

prediction of the future system evolution becomes

possible. Defining adequate strategies is then

feasible. In this paper, we propose our multi-level

model and we detail the first level modeling. The

paper is organized as follows: section 2 presents a

state of the art of multi-agent works studying the

management of renewable and pastoral resources,

section 3 describes our multi-agent model of the

pastoral dynamic using multi-level organisation and

interaction and introduces our first level multi-agent

based model, and we end this work with conclusion

and perspectives.

2 STATE OF THE ART AND

EXPECTED CONTRIBUTIONS

The pastoral system is a complex system

characterized by a large number of entities of

different nature (reactive: plants, animals, cognitive:

shepherds, state). These entities interact with the

environment or with each other to achieve their

goals. In a multi-agent context, some models of

natural resources management focused on modeling

the interaction between human and resources. These

models are based on the stigmergy notion; that is a

change in the environment by an agent affects the

decisions of other agents (Omicini. et al, 2004).

Other models are based on interactions (conflicts,

negotiations, etc.) between agents to make collective

use of resources. In this context, note the role

playing games used to evaluate the impact of

negotiations and decision taken by various

stakeholders on the development of common

resources. In literature, role playing games were

used to study individual and groups behaviors in

economic social context. During such games, each

actor plays the role of the realist stakeholder in a

fictive environment. Role playing games are a

powerful tool to support negotiation process and

participant training (Guyot and Honiden, 2006). In a

multi-agent context, several works tried to jointly

use multi-agent systems and role playing games. For

example, Dray, et al. (2006) introduced the role-

playing game "AtollGame" based on a multi-agent

system to study the problem of drinking water

supplies. The Role playing game "SylvoPast"

proposed by Etienne (2003) studies the negotiation

process between farmers and forestry in order to

prevent fires in Mediterranean forest areas.

Note that most multi-agent systems modeling

natural and pastoral resource management focus on

decisions making and negotiation between economic

and social actors, using variables of macro level (e.g.

hydraulic state of the year: wet year; dry year).

These variables are empirical or dependent on a

number of assumptions. Therefore, such approach

can lead to biased results that can be the cause of

non-adequate decision making. However the

ecological literature is rich on studies of the micro

dynamics of such a system. For example, the study

of hydraulic dynamics provides us with information

about runoff emergence. It is then possible to take

advantages of such information to define with

precision the best shepherd displacement. Such

information is much more interesting then

characterizing the hydraulic state by criteria such as

“dry year” and “wet year”. The study of animals

behavior (imitation, displacement behavior, grazing)

can also be used for space management. For

example, studying the behavior of selectivity

(animal preference) supports reflexions on

introducing mixed herds (with different preferences)

instead of working on criteria such as “dense

vegetation cover” and “naked ground”.

Throughout this work, we opted to go down to

the micro-level modeling of the Tunisian pastoral

dynamics. Taking into account the different levels of

granularity allows best comprehension of the system

dynamic and adequate decision making.

MULTI-AGENT BASED MODELING OF THE TUNISIAN PASTORAL DYNAMIC - Multi-level Organization

255

Figure 1: Pastoral system: Multi-level organization.

Figure 2: Pastoral System: Design and Modeling by organization level.

3 MULTI-LEVL MODELING OF

THE PASTORAL SYSTEM

3.1 Multi-level Organization

In order to manage our system complexity, the basic

idea is to make a multi-level modeling (shown in

Figure 1). This organization will enable us to make

the design and modeling of each level of granularity

in its own (See Figure 2). The model includes five

levels of granularity. The first level of granularity

includes three sub-levels:

• The dynamics of soil affected by climatic

conditions (hydrology, fertility, water

infiltration, runoff).

• The vegetation dynamics (plant growth,

physiological and chronological cycle,

phytosociology).

• The Animal agent behavior (food needs,

preferences).

The interaction dynamic of the soil, vegetations

and climatic variables lead to the emergence of

global properties such as the hydraulic state of the

soil and vegetation density. In order to validate

results at this level, we consider the use of GIS

ICAART 2012 - International Conference on Agents and Artificial Intelligence

256

(Geographic Information System).

The second level of granularity focuses on

animals’ group behavior (imitation, leadership,

competition over resources). The interaction

dynamics in this level of granularity lead to the

emergence of the groups’ displacement and

consumption behavior.

The third level of granularity focuses on

modeling the decisional behavior of the cognitive

Shepherd agent in response to the macro-variables

values provided by the lower level of granularity

(size of the herd, vegetation density).

The fourth level of granularity focuses on

modeling the interactions between Shepherd agents

(negotiation, competition over resources). In order to

stick with reality, we consider the use of

participatory simulation in this level. The simulation

results on the emergence of collective behavior (e.g.

rules of access to resources).

The fifth level of granularity focuses on the

strategic decision level of the State agent defining

the strategies to meet the needs of different

stakeholders while maintaining the sustainability of

the system (grain subsidy, renewal of vegetation

cover).

In this model, the different levels of granularity

are fed by the macro indicators of lower levels (see

Figure 2). Note that indicators are macro variables or

emergent properties resulting of the underlying local

dynamics. The higher levels take into account these

indicators and act by feedback. Taking into account

this feedback, the lower layer dynamics proceed by

adaptation to the occurred changes. In addition to

these indicators, we define for each level of

granularity thresholds. When indicators exceed the

thresholds values, alarming situations are risen (land

cover degradation, soil degradation), so the

underlying higher levels make decision to manage

situations.

3.2 Modeling Multi-level Interaction

To simplify access to data of the different layers of

our model, we opted for the use of an Observer

agent by each layer (See Figure 3). An Observer

agent is characterized by an overall perception of its

layer. It detects the emergence of global properties.

It collects the values of macro indicators, makes

comparison with the thresholds defined to indicate

an alarming situation (as soil degradation). The

Observer agents are charged of the communication

between the different layers. Agents from layer A,

requiring information on a layer B, ask the Observer

agent A for it, which sends a request to the Observer

agent of the layer B. So, this agent responds to the

request. The agents of the A layer act according to

information provided by feedback (flow of action).

When studying interaction between levels of

granularity, we have not taken into account the

notion of time. In fact the appearance of emergent

properties and the establishment of macro variables

require consideration of the underlying dynamics. It

is undeniable that the decision of agents at higher

levels can only be made if there is stabilization in

the variables studied. They subsequently react with

feedback.

Figure 3: Multi-level interactions.

3.3 Micro-level Modeling

In this section, we focus on modeling the first

organization level of our system architecture. Our

efforts are concentrated on soil hydraulic processes

and more specifically the dynamics of infiltration.

To do this, we are inspired by the principle of eco-

resolution (Ferber, 1995). Our basic model consists

of two agents (see Figure 4): Drop of Water agent

and Air Bubble agent. Initially, the Air Bubble agent

is occupying a pore in the soil, The Drop of Water

agent is on the soil surface. The Drop of Water agent

moves to achieve its satisfaction which is occupying

a pore. The Air Bubble agent leaves the pore if it is

attacked. Since we are interested to infiltration

dynamic, we are limited to study the case that. The

Drop of Water agent has a speed which is lower than

the runoff speed. In this case, the Drop of Water

agent attacks close Drop of Water agents. If there is

no Drop of Water agent in its neighbourhood, it

attacks the nearest Air Bubble agent occupying a

MULTI-AGENT BASED MODELING OF THE TUNISIAN PASTORAL DYNAMIC - Multi-level Organization

257

pore forming positive angle with the current position

of the Drop of Water agent.

Figure 4: Infiltration process.

4 CONCLUSIONS AND

PERSPECTIVES

We presented a multi-agent multi-level model of the

Tunisian pastoral system. The multi-level

organization allows best comprehension of the

pastoral dynamics. In terms of development, such

organization offers the possibility to design each

granularity level in its own. The validation of each

level allows moving to the next level of granularity.

Simulation taking into account all levels of

granularity allows the future system dynamics

prediction. This aspect will provide our system

PASDES with the ability to define adequate

strategies taking into account micro- level indicators.

Up to now, we have started the first level of

granularity modeling, we have attacked the soil

hydraulic dynamics and more specifically the

infiltration process. A very important step at this

level is the acquisition of ecological and climatic

data needed to supply our first level models, hence

the need of the collaboration with experts in the

field. The use of GIS for the validation of the

ecological model will allow us to stick well with

reality. Interactions between cognitive agents in

participatory simulation will take advantages of

efficient data provided by this model.

REFERENCES

Bergandi, D., 2000. Eco-cybernetics: the ecology and

cybernetics of missing emergences. Kybernetes 29 (8).

Costanza, R., Wainger, L., Folke, C., Mäler, K. G., 1993.

Modeling complex ecological economic systems:

toward an evolutionary, dynamic understanding of

people and nature. BioScience. 43(8), pp. 545-555.

Dray, A., Perez., P., Jones, N, Le Page, C., D'Aquino, P.,

White, I. and Auatabu, T., 2006. The AtollGame

Experience: from Knowledge Engineering to a

Computer-Assisted Role Playing Game. Journal of

Artificial Societies and Social Simulation, 9 (1).

Drogoul A. and Guyot P., 2004. Multi-Agent Based

Participatory Simulations on Various Scales, In:

Ishida, T., Gasser, L. and Nakashima H., eds.

Massively Multi-agent Systems (MMAS), 2004.

Lecture Notes in Artificial Intelligence, Springer-

Verlag Berlin Heidelberg, pp 149-160.

Dutilly-Diane, C., 2006. Gestion collective des parcours

en zone agro-pastorale : le cas de Ait Ammar (Maroc).

Afrique contemporaine, 3(219), pp.103-117.

Etienne, M., 2003. Sylvopast: a multiple target role-

playing game to assess negotiation processes in

sylvopastoral management planning. Artificial

Societies and Social Simulation, 6(2).

Ferber. J., 1995. Les systèmes multi-agents. Vers une

intelligence collective. InterEditions. Paris.

Guyot, P. and Honiden S., 2006. Agent-Based

Participatory Simulations: Merging Multi-Agent

Systems and Role-Playing Games. Journal of

Artificial Societies and Social Simulation, 9(4).

Lalba, A., Zoundi, J. S., Tiendrebeogo, J. P., 2005.

Politiques agricoles et accès aux parcours communs

dans le terroir de Ouara à l’ouest du Burkina Faso: une

analyse économique et environnementale à l’aide de la

programmation linéaire. Biotechnologie, agronomie,

société et environnement, 9 (1), pp.43-52.

Omicini, A., Ricci, A., Viroli, M., Castelfranchi, C. and

Tummolini, L., 2004. Coordination Artifacts:

Environment-Based Coordination for Intelligent

Agents. In Proceeding of AAMAS, July 2004, pp.286-

293.

Oprisan, S. A., and Oprisan, A., 2006. A computational

model of oncogenesis using the systemic approach.

Axiomathes, 16 (2).

Soares-Filho, B. S., Cerqueira, G. C. and Pennachin, C. L.,

2002. DINAMICA - a stochastic cellular automata

model designed to simulate the landscape dynamics in

an Amazonian colonization frontier. Ecological

Modelling, 154(3), pp.217-235.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

258