This is the html version of the file https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3995855.
Google automatically generates html versions of documents as we crawl the web.

Location via proxy:   [ UP ]   [Report a bug]   [Manage cookies]    No cookies    No scripts    No ads    No referrer    Show this form

Page 1

Title: A survey on knowledge-driven multi-agent computer simulations for disaster management

Keywords: Semantic Web Technologies, Multi-agent Simulation, Knowledge-Driven Architecture,

Disaster Management

Authors: Claire Prudhomme1, Christophe Cruz2, Ana Roxin2, Frank Boochs1

1. Institut i3mainz de l’Université des Sciences appliquées, Lucy-Hillebrand-Str. 2, 55128 Mainz,

Germany {claire.prudhomme, frank.boochs}@hs-mainz.de

2. Laboratoire d’Informatique de Bourgogne (LIB) - EA 7534, University of Bourgogne Franche-Comté,

Aile des Sciences de l'Ingénieur, 9 avenue Alain Savary, BP 47870, 21078 Dijon CEDEX, France

{christophe.cruz, ana.roxin}@ubfc.fr

Abstract: Protecting humans from disasters has been an active mission of governments and experts

through the definition of disaster management plans. Defining disaster response strategies are crucial

to reduce the number of victims and the economic impact. An efficient preparedness depends on the

cycle which aims at planning and organizing disaster response, experimenting the plans through

training and exercise, and Assessment of the planning based on the experiments. Therefore, plan

assessment is an essential step in the preparedness to identify potential problems and planning errors.

However, the plan assessment requires a plan experiment step. There are two ways to experiment

plans: exercises and computer simulations. This survey focuses on preparedness phases for plan

experimentation and effectiveness assessment by providing details about last work on mulit-agent

simulation methods and knowledge engineering approaches. This survey answers the following

question : how do multi-agent simulations and semantic representations of disaster management

provide an effective way to assess disaster management plans? How to measure the quality of disaster

management plans? Indeed, the plan’s effectiveness must be measured to support its improvement.

1 Introduction

Disaster management is an infinite cycle that continuously improves the resilience and efficiency to

face disasters. There are some variations in the number and the disaster management steps' names

among the disaster management community. These variations are mainly due to diverse levels of

description. Our presentation follows the most widespread vision of this cycle presented in [Coppola,

2011], [BBK, 2015] and which consists of four steps: mitigation, preparedness, response, and recovery.

Figure 2.1 illustrates this cycle.

Figure 1. The cycle of disaster management

The Mitigation step consists of the identification of risks. According to [Coppola, 2011], the risk is

assessed according to the hazard likelihood and the hazard impact. The hazard corresponds to a

Electronic copy available at: https://ssrn.com/abstract=3995855

potential disaster. Its likelihood is often calculated according to the historical events of this hazard.

The calculation also considers the changes, which could increase or reduce this hazard. The impact

depends mainly on the vulnerability of the population and the infrastructure. The risk assessment aims

at reducing the exposure of the population by prevention. The Mitigation also aims at lowering

infrastructure vulnerability. This vulnerability reduction comes from the creation of the norms (e.g.,

anti-seismic norms) and structural reinforcements. Its goal is to avoid disaster or at least reduce its

impact.

The Preparedness step precedes the disaster event. This step aims to elaborate on how to react and

act in front of disaster impacts. This step’s first activity is planning a response that implies using a risk

analysis system to create a plan adapted to potential disasters. Each disaster is decomposed into

different severity levels to identify the different needs according to its severity level. The second

activity corresponds to the preparation of equipment and resources which have been identified in the

action plan. The equipment and resources have a determining role in the disaster; if they are missing,

the consequences can severely impact human lives. The third activity is the actors’ response training

according to their role and actions required during a disaster — some examples of training concern the

evacuation, the management of volunteers, or injuries. This training aims to learn how to act, but the

learning is not enough, and practice is required. Consequently, the fourth activity is the exercise, which

aims to train the population and the responders and assess plans. The last activity is the monitoring of

the potential hazards to early warning. Training, exercises, resource preparation, and early warning

depend on the first activity, which is the conception of plans. It requires an overview of the risks and

available means for stakeholders, who have specific capacities, and resources.

The response step begins immediately when a disaster happens or is imminent and corresponds to

emergency or crisis management. Its goal is to limit the number of casualties, damages, and impacts

on the environment. The response step requires evaluating the situation: what areas are affected,

what type and number of people have been affected, what damages, what needs are present, and

what organizations, resources, and equipment are available to act. Besides, this step requires to use

of situation analysis techniques to identify solutions to be applied. The management between the

situation needs and capacities provided by the different actors of the response remains a difficult task.

This task consists of decision-making to manage resources and solve problems resulting from a

disaster. The collaboration between the different actors plays a crucial role in the response efficiency.

Without cooperation, actors would lose precious time.

Two phases compose the recovery step, short and long terms. The first one consists of providing the

necessary to live for each affected people. Humanitarian aid organizations that also consist of logistic

work to retrieve a minimum of communication networks manage the phase. The second phase consists

of returning to a normal situation. The achievement requires an evaluation of the loss and damages

caused by a disaster. This evaluation allows then cleaning, temporary sheltering of population, and

rebuilding. These three stages require an organization and planning to be as efficient as much as

possible. Moreover, rebuilding gathers many constraints like time, cost, and improvement according

to past weaknesses.

The response phase is the most critical in terms of time and challenge since human lives depend on

this phase's effectiveness. This effectiveness depends on the collaboration between the different

stakeholders. The collaboration has two aspects: coordination and cooperation [Gulati et al., 2012].

• Coordination is the act of managing interdependencies between activities performed to achieve a

goal [Malone and Crowston, 1990]. The coordination components are activities that correspond

Electronic copy available at: https://ssrn.com/abstract=3995855

to the goal decomposition, actions selected and assigned to actors, and interdependency, which

must be managed [Malone and Crowston, 1990].

• The coordination aspect is typically the task carried out during the preparedness. It corresponds

to the planning task to organize actions between the different stakeholders. The cooperation is

defined as "joint pursuit of an agreed-on goal(s) in a manner corresponding to a shared

understanding about contributions and payoffs" [Gulati et al., 2012].

Preparedness is essential to provide a shared view during the preparedness and response phases.

Coordination and cooperation make use of information and knowledge sharing. However, the study of

the literature and field exercises made by [Bharosa et al., 2010] highlights the challenges and obstacles

in sharing and coordinating information that reduces disaster response effectiveness. During a

multiagency disaster response, collaboration obstacles appear at community, agency, and individual

levels. Although plans exist for each of these levels, some organizational silos, conflicting role structure,

a mismatch between goals, allocation of responsibility, or inability to determine what should be shared

show a lack in the preparation process. Indeed such problems appear during the response step and

impact its efficiency, but they must be carried out upstream during the preparedness step.

An efficient preparedness depends on the cycle which aims at (1) planning and organizing disaster

response, (2) experimenting the plans through training and exercise, and (3) Assessment of the

planning based on the experiments. Therefore, plan assessment is an essential step in the

preparedness to identify potential problems and planning errors. However, the plan assessment

requires a plan experiment step. There are two ways to experiment plans: exercises and computer

simulations.

• Exercises aim at testing and improving the collaboration between the different stakeholders. It

aims at simulating the real condition of a disaster according to a scenario. The practices gather all

stakeholders which intervene in disaster response and have a high cost to organize them, which

makes them occasional. The high cost of exercises limits their number and the possibility to test

scenarios essential in the response plan's assessment process.

• Computer simulations aim at testing designed plans or different strategies of action. Compared to

exercises, computer simulation's main advantage is to allow a high number of experiments to

assess a model or a strategy. Computer simulation is a low-cost and effective method for testing

multiple inputs and assessing different outputs to observe a system [Brown and Robinson, 2005].

Computer simulations gather several types of techniques. It is essential to choose the simulation

technique according to the simulation goal and the system to model. The authors of [Mishra et al.,

2019] review the simulation techniques used in the literature for disaster management and examine

the issues they address. The more commonly used identified techniques are discrete-event simulation,

system dynamics, agent-based simulation, and Monte Carlo simulation. The examination of the issues

addressed by these techniques shows the agent-based simulation as the most used during the

preparedness to assess strategies and plans of disaster response [Mishra et al., 2019].

The usage of agent-based simulation during the real-time Response is minor compared to the use

during the preparedness. Multi-agent simulation is an appropriate technique to achieve the practical

application of plans in diverse situations. Such a method requires modeling and design steps to

produce simulation experiment results. These results are the basis for studying the achievement of the

plans’ goals and evaluate their effectiveness. However, simulating the diversity of plans and situations

implies a change of goals, stakeholders, actions, and interactions. These changes indicate an

adaptation of the practical application through simulation and an adaptation of effectiveness

assessment. The granularity of multi-agent simulations allows the representation of disaster

Electronic copy available at: https://ssrn.com/abstract=3995855

management stakeholders, their organizational structure, their interactions, and their actions that

come from their preparation and knowledge according to the situation and population behavior

according to their specificities.

Among the usage during preparedness, most simulation approaches focus on determining optimal

solutions to a problem using an agent behavior based on reality rather than focusing on the elaboration

of agent behaviors using new algorithms. The agent behavior based on reality corresponds to the

assessment of existing strategies to identify the best one when there are several possibilities or the

best parameter for applying an approach.

This survey focuses on preparedness phases for plan experimentation and effectiveness assessment

by providing details about last work on mulit-agent simulation methods and knowledge engineering

approaches. This survey answers the following question : how do multi-agent simulations and semantic

representations of disaster management provide an effective way to assess disaster management

plans? How to measure the quality of disaster management plans? Indeed, the plan’s effectiveness

must be measured to support its improvement.

The second section presents the application domain which is the disaster management, to firstly

highlight the business problem of plan assessment and secondly, highlight the limits of existing

approaches related to the plan assessment. The third section reviews engineering techniques,

components, and platforms for multi-agent simulations, which are the most suitable approaches to

experiment and assess disaster management plans. These reviews aim to search suitable approaches

and identify lacks of existing approaches to reach the objectives defined in the first section. The fourth

section presents the knowledge engineering domain to allow the representation of the disaster

management knowledge, whose plans’ description. It firstly reviews the existing ontologies of this

domain and secondly reviews approaches to integrate knowledge extracted from data. Finally, the

chapter concludes this article.

Electronic copy available at: https://ssrn.com/abstract=3995855

2 Preparedness

The preparedness step aims at preparing the collaboration between the different stakeholders of

disaster management. This collaboration must be anticipated through activities and responsibilities

planning to coordinate stakeholders. The Federal Emergency Management Agency (FEMA)1 [Federal

Emergency Management Agency, 2010] presents the preparedness step as a cycle of five sub-steps:

Plan, Organize/Equip, Train, Exercise, Evaluate/Improve. Figure 2 illustrates the cycle of preparedness.

The heart of the preparedness is the planning sub-step that guides the other sub-steps.

Plan

Organize

Equip

Train

Exercice

Evaluate

Improve

Figure 2. Cycle of preparedness according to [Federal Emergency Management Agency, 2010]

Planning aims at managing risks by considering all hazards and threats. It must take into account the

whole population and its needs to provide an adapted solution. It should also be flexible enough to

address both traditional and catastrophic incidents. Planning must be a collaborative process between

all community stakeholders to identify the missions and support goals. Planning identifies tasks,

allocates resources to accomplish those tasks, and establishes accountability, according to a

description of the anticipated environment for action.

Such collaborative preparation makes Preparedness crucial for disaster management. The entire

structure of this phase stands on the development of plans. Preparedness aims at ensuring a

preparation adapted to needs. In disaster management, there are several needs: a need for an

organizational structure, a need for risk management, and a need for situation management.

Addressing such needs results in different levels of plans. These levels of plans allow providing a

common and homogeneous structure between different administrative areas. They aim to facilitate

their collaboration and define a specific preparation adapted to both the needs of each of them and

the disaster severity. These different levels of plans are detailed in the following subsection.

2.1 Plan preparation

The phase of planning consists firstly of determining the general needs of disaster management in

terms of organizational structure (e.g., command post, responsibilities), tasks (e.g., informing the

population, closing roads, build walls against water), and roles (e.g., communication management,

security management, flood protection management). Secondly, it is necessary to plan the disaster

management operations and their coordination according to levels of risk (e.g., orange and red risk

levels, which are the higher risk levels of the four French risk levels) specific to an administrative area.

The last step is to plan the implementation of operations adapted to the current disaster situation.

These three steps correspond to planning at three levels of detail: strategic, operational, and tactical.

1 Federal Emergency Management Agency (FEMA): https://www.fema.gov/, visited on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

• Strategic plans are guidelines and general plans defining a standard process and a standard

structure for a jurisdiction. This standard process and structure aims to facilitate the

collaboration between different administrative areas through a standard model of disaster

management while allowing a plan design specific to each of them and the severity of a

disaster. This type of plan is defined by [Federal Emergency Management Agency, 2010] as

following: "Strategic plans describe how a jurisdiction wants to meet its emergency

management or homeland security responsibilities over the long-term. These plans are driven

by policy from senior officials and establish planning priorities." A strategic plan defines the

general tasks to prepare and achieve in case of disaster, the roles, and the responsibilities

associated with these tasks. Besides, it provides an organizational structure by defining

command centers and a hierarchy of managers. A commander's role is generally at the charge

of an administrative responsible or an executive officer of an organization playing a crucial role

in disaster management. For example, a mayor can be defined as the director of rescue

operations. The chief of fire brigade can be defined as the rescue commander for a disaster at

a municipality level.

• Operational plans are plans prepared by each administrative area to address specific needs.

This planning depends on the means of the administrative area and must respect the strategic

plans of its jurisdiction. This plan aims at organizing the collaborative work between the

different stakeholders of disaster management. This type of plan defines people and

organizations associated with a role, details tasks, and actions to perform by a function to

address risks of the locality. They also contain an estimation of the resources required for their

achievement. The locality inventory of resources that are available in case of disaster

accompanies the operational plans. An operational plan is defined by [Federal Emergency

Management Agency, 2010] as following: "Operational plans provide a description of roles and

responsibilities, tasks, integration, and actions required of jurisdiction or its departments and

agencies during emergencies. Jurisdictions use plans to provide the goals, roles, and

responsibilities that a jurisdiction’s departments and agencies are assigned and focus on

coordinating and integrating the many responses and support organizations within a domain.

They also consider private sector planning efforts as an integral part of community-based

planning, and to ensure efficient allocation of resources. Department and agency plans do the

same thing for the internal elements of those organizations. Operational plans tend to focus

more on the broader physical, spatial, and time-related dimensions of operation; thus, they

tend to be more complex and comprehensive, yet less defined, than tactical plans." The

operational plans depend on risk estimation. They coordinate actions and tasks according to

specific events (e.g., a certain water level during a flood) or a particular warning risk level.

These tasks can be described by other plans designed by an organization in charge of the task.

Moreover, stakeholders (people and organizations), resources, tasks, and actions are located.

• Tactical plans are the planning done during the response, corresponding to the resources

management and detailed planning based on operational plans activation to respond

accurately to the situation. A tactical plan corresponds to the decisions made to manage a

disaster scenario based on the preparation. These plans aim at defining precisely who

intervene, what, where, and how to respond to the disaster situation. Their goal is to respond

accurately to the situation needs according to the responsibilities of the stakeholders. They

define a protocol of actions and manage resources (e.g., humans, vehicles, equipment) to

achieve a task according to the situation conditions (e.g., events, weather, location, available

resources). A tactical plan is defined by [Federal Emergency Management Agency, 2010] as

following: "Tactical plans focus on managing personnel, equipment, and resources that play a

direct role in incident response. Preincident tactical planning, based upon existing operational

Electronic copy available at: https://ssrn.com/abstract=3995855

plans, provides the opportunity to pre-identify personnel, equipment, exercise, and training

requirements. These gaps can then be filled through various means (e.g., mutual aid, technical

assistance, updates to the policy, procurement, contingency leasing)."

These three levels of plans are defined by the Federal Emergency Management Agency (FEMA) of the

United States in [Federal Emergency Management Agency, 2010]. Still, they are not specific to the

United States. Indeed, the authors of [Labba et al., 2017] study the organizational structure of disaster

management in the United States, the United Kingdom, and France. They observe these three levels

in common to each of these countries. The European project EPISECC also highlights these three levels

of management in [Information, 2014]. However, in Europe, there are variations in the definition of

operational and tactical levels. For example, in France, definitions of operational and tactical levels are

inverted compared to the previously presented description.

• In France, they use the term tactical level to describe administrative staff and commander that

make decisions, but who are not on the ground. They use the term operational level for the

management on the ground. This difference is due to the meaning of operations, which are actions

directly performed on the ground.

• In Germany, operational and tactical levels work together to manage operations through

command and control management thanks to the proximity between the command center and

people on the ground.

• In the United States, the tactical level is nearest to the ground view because it means technically

achieving an operation.

Despite the variations of the definition and the structural organization, the three aspects are present

with a first standard level that provides organizational structure and guidelines to plan disaster

management (corresponding to strategic aspect); and a second level with operational and tactical

elements, generally leads by a responsible of jurisdiction.

2.1.1 Plan Experiment

The real exercises require reproducing disaster scenarios and gather all stakeholders to simulate and

test the plans’ implementation according to real situations. It involves the setup of a lot of means and

money to reproduce disaster scenarios. More exercises need to be realistic, and more costs are high.

The exercises have the advantage of allowing both the assessment of the plan’s effectiveness and

stakeholders’ training simultaneously. Although there are a few biases during an exercise due to the

responder’s mindset that fake situation, but real exercises are quite precise. Its primary disadvantage

is a high organizing cost to gather all stakeholders and set up the most real disaster situations.

Computer simulation permit to assess a system through experiments. For the goal of plan assessment,

experiments can simulate the plan-based disaster management system in a broad diversity of

scenarios. Such simulation requires a system model expressed through informatics paradigms.

"A model is a representation of a system that can be defined and studied

indirectly by helping to provide answers about it". [Rodrigues Da Silva, 2015]

Three design steps constitute the simulation modeling process:

• the design of a conceptual model (i.e., model conceptualization);

• the design of the communicative model (i.e., representation of the conceptualization independent

of a platform);

• the design of the programmed model (i.e., programming of the model);

• the design of the experiment model (i.e., experiment set up).

Electronic copy available at: https://ssrn.com/abstract=3995855

According to [Nance, 1994], the conceptual model is "that model which exists in the mind of the

modeler." In contrast, the communicative model is a model representation that can be communicated

to others. In the literature [Benjamin et al., 2006], the distinction between the conceptual model and

its expression (i.e., the communicative model) is not always clear. Therefore, the term conceptual

model is more often used to describe its expression than its abstraction in the designer’s mind.

The programmed model also called the computational model, "is a model representation that admits

execution by a computer to produce simulation results" [Nance, 1994]. Finally, the experimental model

is formed by adding executable descriptions of the test environment to the programmed model that

allows the simulation execution. It makes use of different parameter configurations allowing the

simulation of the system model in different scenarios. Computer simulation has the advantage of being

a low-cost approach for testing a plan in a broad diversity of scenarios. Although good realism,

simulations can be less precise than real exercise.

A proper evaluation of plans needs to be based on several exercises to provide extensive feedback to

test strategies addressing the situation and evaluate them. Nevertheless, the high cost of real exercises

limits the number of iterations, the complexity, and the realism of scenarios. Although simulations

have less precision than real exercises, but the simulation precision can reach enough high precision

to assess the plan’s effectiveness. Therefore computer simulations are the most suited for practical

plan applications to evaluate their effectiveness. Thus, computer simulations provide plan testing

capability in a large and complex variety of scenarios for a low cost.

2.1.2 Plan Effectiveness Assessment

The evaluation of a plan’s effectiveness should allow determining the conditions under which a plan

can be applied, and its success rate under these conditions. To this end, it is necessary to define the

effectiveness of a plan quantitatively using metrics. However, the effectiveness of a plan depends on

its objective’s success, which can affect various areas. The "Sphere project" [The Sphere Project, 2011]

has identified the following areas: health, shelter, food and nutrition, water, and sanitation. It is,

therefore, necessary to define metrics to evaluate plans in these different areas. These metrics help to

group situations for which a plan has similar effectiveness to extract the common characteristics of

these situations. These characteristics allow the identification of the applicability conditions of a plan.

Thus, plan assessment requires clustering on simulation data to group simulations for which the plan

has similar effectiveness. This is a data clustering problem.

Several metrics have been developed to support plan assessment. In [Larsson, 2008], the authors

identify three main criteria for assessing plans, including the effectiveness criterion. The authors

propose to evaluate the effectiveness of a plan according to the following parameters:

1. Victims found

2. Victims whose condition worsens

3. Identified property domage

4. Property sustained further domage

5. Infrastructure operating

These parameters allow the evaluation of the plans’ effectiveness in the field of health and shelter. In

[Bayram et al., 2012], the authors identify 12 quantitative parameters to make an assessment based

on health, shelter, food and nutrition, water, and sanitation. These 12 quantitative parameters are a

basis for assessing the plans in terms of the areas they address:

1. Number of Excess Deaths

2. Number of Under-5 Excess Deaths

Electronic copy available at: https://ssrn.com/abstract=3995855

3. Number of Cases with Acute Communicable Diseases

4. Number of Cases with Traumatic/Chemical/Radiological Injuries

5. Level of Health Care Services

6. Number of Young Children (6-59 months) with Acute Malnutrition

7. Number of Displaced Persons (internally displaced or refugees)

8. Number of Persons with Inadequate Living Space

9. Water Quantity

10. Water Quality

11. Level of Sanitation Facilities

12. Gender-based violence

The metric used to compute the effectiveness of a plan must be adapted to its goal. The plan’s goal is

linked to one or several of the presented quantitative parameters. Therefore, the effectiveness metric

must be computed with the parameters related to the plan’s goal. These parameters are observed

variables during a simulation. In addition to computing the effectiveness metric, it is essential to define

its associated applicability context. Such a definition requires gathering simulation with similar

effectiveness and identifying the criteria that characterize the effectiveness’ applicability context.

When assessing plans, clustering is carried out on various dimensions, where each dimension

represents a criterion. However, criteria impact a plan’s effectiveness differently, and some criteria

may not impact the plan’s effectiveness. Unsupervised clustering by considering criteria that do not

impact the plan’s effectiveness leads to oversegmentation of groups. Therefore, it is necessary to

identify the criteria that do not have a significant impact on the plans.

The review [Saxena et al., 2017] presents most relevant approaches for Unsupervised Data Clustering.

This review splits the different approaches in clustering approaches in two families: Hierarchical

clustering and Partitional clustering. Hierarchical clustering can be agglomerative or divisive and use

single, complete, or average-linkage. The studied approaches of this family are the approaches BIRCH

[Zhang et al., 1996], CURE [Guha et al., 2001], ROCK [Guha et al., 2000], and CHAMELEON [Karypis et

al., 1999]. Partitional clustering is divided between distance-based, Model-based, and density-based

approaches. Distance-based approaches use Error Square, whereas model-based and density-based

approaches use probabilistic. The studied approaches for the Partitional clustering family are the

approaches K-means [MacQueen et al., 1967], CLARANS [Ng and Han, 2002], FCM [Dunn, 1973].

Among the different approaches, the hierarchical clustering CURE is the most relevant approach for its

low complexity (i.e., O(n2log(n)) as worst-case time complexity), its high scalability, and its suitability

for large data and low sensibility to outliers. Although BIRCH approach has a better time complexity

than CURE approach, CURE approach has better quality than BIRCH approach. Moreover, BIRCH is not

suitable for high-dimensional data. Therefore, the CURE approach appears as the best compromise

between time complexity and clustering quality.

2.1.3 Discussion

Preparedness is the essential step of disaster management to guaranty an effective response. It aims

at preparing plans, assessing them, preparing resources and stakeholders by training. The assessment

of plans plays a crucial role in guarantee good preparedness and effective response. Indeed, being

prepared and trained to apply an ineffective plan is useless; that is why the assessment of plans is

essential to know its effectiveness. Therefore, the plan’s effectiveness must be measured to support

its improvement. Although some metrics are proposed in the literature and presented previously,

there is no standardized system of measurement necessary for comparison [Guha-Sapir and Below,

2002]. A method to measure the plan’s effectiveness must identify objectively, deficiencies in the

application of plans. It thus requires to experiment plans to assess their global and their case-specific

Electronic copy available at: https://ssrn.com/abstract=3995855

applicability through low-cost simulation. The assessment requires thus clustering approach to gather

plans according to their common specificities that make them effective. Low-cost simulations must

experiment plans to allow such an analysis. Therefore, computer simulations are the most suited to

provide a vast diversity of experiments with the lowest cost. The next section presents the simulation

approaches for disaster management.

2.2 Simulation approaches supporting preparedness

Simulation techniques are often used in the disaster community to support them in decision-making.

They are mainly used to assess risk through disaster simulation or to assess response strategies during

the preparation or in real-time during the Response. The review of [Mishra et al., 2019] on disaster

management simulation modeling highlights the use of four main techniques: System dynamics,

Monte Carlo simulation, discrete-event simulation, and agent-based simulation. They analyze the

repartition of the different methods according to disaster management topics and disaster

management steps. The system dynamics are the most present (42%), followed by Monte Carlo

simulation (25%), followed by agent-based simulation (22%), and finally, discrete-event simulations

(11%). System dynamics are mainly used during Mitigation for risk assessment/identification (21%) and

vulnerability assessment (4%). It is also used during preparedness for prevention and recovery

schemes (15%). Monte Carlo simulations are used during Mitigation for risk modeling (16%). Also, they

are sometimes used during Response for solving disaster relief (2%). Agent-based simulations are

mainly used during Preparedness and Response for simulating rescue and evacuation strategies (13%),

disaster management (4%), and modeling healthcare (2%). Finally, discrete-event simulations are

mainly used during preparedness and response to model large-scale (8%). Among the different

simulation techniques used for disaster management, the agent-based simulation, also called multi-

agent simulation, is the most used during the preparedness to assess strategies and plans of disaster

response [Mishra et al., 2019]. The granularity of multi-agent simulations allows the representation of

disaster management stakeholders, their organizational structure, their interactions, and their actions

that come from their preparation and knowledge according to the situation, but also population

behavior according to their specificities. This technique allows the simulation of complex systems

without modeling it directly but by obtaining it by the emergence of its model components and their

interactions. That is why the simulation technique based on a multi-agent system is the most suited

simulation technique to address the plan experiments.

According to four different usages, the survey [Hawe et al., 2012] presents a classification of agent-

based simulation for large-scale emergency response. The first category of usage (called U4 in [Hawe

et al., 2012]) corresponds to real-time Response simulations. The three other categories belong to

simulations for preparedness. The second category (called U3 in [Hawe et al., 2012]) corresponds to

preparedness simulation with agent behavior using new algorithms. The two last categories of

preparedness simulations use agent behavior based on reality. The third category (called U2 in [Hawe

et al., 2012]) corresponds to preparedness simulation that searches for a specific optimal response.

The fourth category (called U1 in [Hawe et al., 2012]) corresponds to preparedness simulation that

searches for a generalized optimal response. The usage of agent-based simulation during the real-time

Response is minor compared to the usage during the preparedness. Among the usage during

preparedness, most simulation approaches focus on determining optimal solutions to a problem using

an agent behavior based on reality rather than focusing on the elaboration of agent behaviors using

new algorithms. The agent behavior based on reality corresponds to the assessment of existing

strategies to identify the best one when there are several possibilities or the best parameter for

applying an approach.

Electronic copy available at: https://ssrn.com/abstract=3995855

The majority of simulation approaches searching for optimized solutions are designed for a specific

problem [Hawe et al., 2012]. The most addressed specific issue is the allocation of resources both in

terms of human rescuers [Praiwattana and El Rhalibi, 2016] or robotic rescuers [Blatt et al., 2016], and

of resource means [Hawe et al., 2015], [Marecki et al., 2005]. For the optimization of allocated

resources and the planning of response action, the simulation results are assessed in terms of time

and success quantity (e.g., ratio rescued people) according to the purpose of the strategy assessed by

the simulation or the use case addressed by the simulation. This type of simulation addresses mainly

the rescue strategies. Another set of simulation focuses on evacuation simulations [Christensen and

Sasaki, 2008], [D’Orazio et al., 2014], [Zhou et al., 2012], [Mas et al., 2015], [Nagarajan et al., 2012] by

taking into account some aspects as the behavior of the population, the traffic, communications, and

prepared plans. Other approaches that focus on the generalized optimal solution as [Saoud et al., 2006]

experiments and compare results obtained by different strategies. The combination of varying action

strategies corresponds to different plans. Therefore, such general optimization approaches assess

according to specific criteria a plan to define the optimal one according to some situation’s

characteristics.

However, all of these simulation approaches for optimal solutions are specialized for a problem, which

is more or less specific. The specificity of their design ranges from identifying the optimal parameters

of a plan (meaning configuring a plan optimally) for a specific situation corresponding to an optimal

solution. The specificity of the simulation model, according to a problem, is essential to simulate and

assess a plan accordingly. However, the limit of the existing simulation approaches is their design and

experiment process. The simulation models and their implementations are limited to use case and a

predefined set in plan variations. This case-dependent design process limits the adaptation and the

extension of these approaches to the diversity of disaster situations and the variety of response plans.

For example, the approach of [Saoud et al., 2006], which is "generic per accident and location, "

assesses the impact of some variations in the NOVI plan, which aims to rescue a large number of

victims. This approach could be extended to allow the study of further variations of this plan based on

a specific organizational structure. However, the assessment of another rescue plan based on another

organizational structure would require a new design process of simulation to define a new model and

new implementation. Through simulations, plan assessment requires modeling and implementing

several types of agents and several organizations corresponding to groups of agents to represent all

stakeholders, their different behaviors, and their different organizational structures. Therefore, the

first limit of existing approaches for evaluating disaster management is the adaptability of simulation

models and experiments to the diversity of disaster management plans. The process of modeling

simulation and experiments is subjective. Thus, the first limit implies a risk of including different biases

during the modeling of simulations and experiments for different plans. These different biases can

compromise their evaluation and comparison.

Approaches of multi-agent simulation engineering presented propose metamodels and ontologies

combined with specific system architecture allowing generative programming. They aim to solve the

lack of flexibility and reusability of simulation models and their implementation in the same application

domain. The study of these approaches shows the benefits of using ontologies to represent a

simulation metamodel and a system architecture based on an extendable and adaptable agent’s

behavior to provide flexibility and reusability in simulation modeling and experiments. The

combination of modeling with a set of implemented multi-agent system components as made in

[Poveda et al., 2015], [Boufedji et al., 2018] allows the extension and reusing components of the

implementation model to simulate a variety of contexts.

Electronic copy available at: https://ssrn.com/abstract=3995855

The implemented components of multi-agent systems depend on the used simulation platform (e.g.

AGLOBE, Repast, Gama, etc.). Such engineering approaches provide flexibility and reusability of multi-

agent components for model variations. This study shows that the adaptation of simulation modeling

and experiments requires a flexible implementation based on the combination of a set of multi-agent

system components. However, existing approaches lack to be reused for disaster management plans

assessment through simulation experiments. Therefore, the second limit is the adaptation and

reusability of implemented multi-agent simulation components to allow a diversity of implementing a

simulation model for plan assessment. This thesis's objective to face this second limit is to provide and

use a set of implemented components that can be extended and reused to allow processing the

diversity of plans. Such an objective requires reviewing the components of existing multi-agent

simulations for disaster management, presented in the next section to identify generic and specific

components of the different approaches. It also requires identifying a suitable simulation platform

allowing such implementation and extensibility. The study of the suitable platform is presented in the

next section.

Moreover, these approaches highlight the third limit of existing approaches for plan assessment,

limiting expressivity. Indeed, multi-agent approaches for modeling disaster management simulation

gathers knowledge about plans and disaster scenarios from experts ad hoc to model them in the multi-

agent paradigm. Thus, existing approaches are addressed for the computer expert community to

model disaster management simulation in a multi-agent paradigm in cooperation with the domain

experts. However, they are not addressed to a disaster management community that searches at

assessing different plans to prepare them to face disaster. Therefore, the third limit concerns the

expressivity and interoperability of plan representation for disaster management communities to

gather their knowledge and assess plans according to their definition of plans and knowledge. Such

objective relates to the domain of knowledge engineering, whose related work is presented in Section

4. Table 1 summarizes the identified limits of plan assessment approaches and objectives to face them

linked to the next sections.

Limits

Objectives

Study

(1) Adaptability of simulation

models and experiments to the

diversity of plans

representation

Increase Flexibility to allow the

assessment of the diverse

plans of disaster management

through their simulation

The domain of multi-agent

simulation engineering section

3.1

(2) Adaptability and Reusability

of MAS components for

disaster management

Increase Extensibility by

allowing to process the

diversity of plans

Features of existing multi-

agent simulations in section

3.2 and simulation platform

allowing such extensibility in

section 3.3

(3) Expressivity of system for

plans and scenarios

Increase Expressivity and

Interoperability to gather DM

knowledge and allow DM

community to assess plans

according to their plan

definition

Domain of knowledge

engineering in section 4

Table 1. Limits of existing multi-agent simulations for plan assessment

3 Multi-agent Simulation

The previous section has presented the benefits of simulation to experiment plans and assess them. It

has shown that the multi-agent-based technique is the most suited for plan experiments through

Electronic copy available at: https://ssrn.com/abstract=3995855

simulation. It has finally highlighted the limits of existing work in the adaptability and reusability of

both simulation models and components. Therefore, this section reviews in the first subsection, the

existing work to adapt the simulation model and facilitate the simulation development to the diversity

of disaster management plans. This review identifies some common points to facilitate the

development of simulation from a variety of models. However, existing approaches are not flexible

enough to be used for the plan assessment. Thus, the second subsection presents a review of

components used in the multi-agent simulation for preparedness to identify the most suited

components of a simulation model for Preparedness. Finally, it reviews simulation platforms in the

third subsection to determine the most adapted one to fit the simulation components' requirements

and flexibility.

3.1 Multi-agent Simulation engineering

As part of disaster management preparedness, computer simulations are often used to optimize

various response plans and determine the tasks, actions, or resources best suited to a situation. Among

existing approaches, those based on multi-agent systems are the most widely used in this context to

simulate stakeholders' behaviors and interactions [Mishra et al., 2019]. It has to be noted that disaster

management stakeholders and their behavior varies from one locality to another. They also vary within

the same locality depending on the disaster situation and needs. A simulation model's definition

adapted to different organizations and plans implies variability in entities, behaviors, and interactions.

However, multi-agent simulation approaches are usually created with a specific objective in mind,

limiting their reuse and explaining the large number of models in this field. As pointed out by the

authors of [Poveda et al., 2015], the lack of interest in sharing or connecting the work done condemns

researchers in this field to reinvent the wheel by creating new simulation models and new

implementations of these models to run the simulations. However, there is some work dealing with

the adaptation of multi-agent simulations for disaster management.

• Among these approaches, the approach presented by the authors of [Poveda et al., 2015] presents

a general model for the design of emergency management services in indoor environments. This

model comprises three layers: a semantic layer, a simulation layer, and a layer containing the

components of the emergency service. The latter layer is based on the simulation layer and has to

be adapted to the context. To solve the problem of reusing and adapting various contexts, the

authors of [Poveda et al., 2015] used a semantic layer composed of the Einsim ontology that links

external data and guides agent-based simulation. The simulation process is realized through a

simulation control system that includes semantic representations in a model repository of

emergency service components and an adaptation model that maps the agent code's semantic

emergency metadata. Representing the simulation process of emergencies through an ontology

allows simulations to be modeled independently of their programmed model. This advantage

provides flexibility in extending and reusing the programmed model to simulate a variety of

contexts. However, being specialized for social simulation in an indoor environment, this ontology

is not general enough to simulate disaster management plans that are not limited to an indoor

environment. Compared to emergency planning in an indoor environment, the simulation of

disaster management plans requires a more complex representation with the organizational

structure of actors, resource management, and a wide variety of plans.

• Another interesting modeling approach for disaster emergency preparedness is presented in

[Kruchten et al., 2007]. The authors present a conceptual model (formalized by an ontology) of

disasters affecting critical infrastructure (energy, transport, communication, etc.). A conceptual

model is an abstraction of the essential characteristics of the studied system. This conceptual

model describes four main components and their interactions, disaster events, infrastructures,

agents involved in disaster management, and the impacts of the events on the population.

Electronic copy available at: https://ssrn.com/abstract=3995855

Disasters affect the well-being of the population and the condition of infrastructure elements. The

latter is the object of observation by the agents and is the target of their actions. This conceptual

model has the advantage of representing at a high level of abstraction and the main components

of disaster management plans using concepts of the agent paradigm. This model provides a

common language for communicating, analyzing, and simulating the interdependencies of critical

infrastructures. The simulation of these interdependencies is intended to detect potential

problems in the plans. This objective explains the simulation model's lack of description because

it does not allow the definition of the simulation model’s variables and goals. Indeed, this model

limits the use of simulation to the study of potential problems in plans.

• Among the non-specific approaches to disaster management, the work of [Boufedji et al., 2018]

presents the adoption of variability models to describe the generic and specific aspects of a multi-

agent system model. The generic aspects of the model are represented according to the agent

paradigm's concepts such as agent, environment, interaction, and organization. In contrast, the

specific aspects are represented according to the specific application domain. These models are

respectively related to the implementation of reusable, generic, and specific multi-agent system

components. Although this work does not address specific aspects of simulation modeling (e.g.

model variables, experimentation), it provides a method for reusing multi-agent system

components for various applications represented by various models.

• Another interesting approach not specific to the disaster management field but specific to the field

of multi-agent simulation is presented in [Christley et al., 2004]. This work presents an ontology

for automating agent-based modeling and simulation tasks. This ontology describes the different

models involved in simulation design (e.g., conceptual, experimental, programmed models), the

basic concepts of the agent paradigm (e.g., agent, environment), as well as concepts linking with

simulation experiments (e.g., simulation data), and the programming of the simulation required

for its execution (e.g., software programming). This ontology is suitable for any multi-agent

simulation modeling and design, whatever the field of application. It offers an ideal set of concepts

for representing various simulation models, experiments, and their development as executable

programs.

Approaches

Model

Agent

Environment

Action

[Christley et al., 2004]

X

X

X

X

[Kruchten et al., 2007]

X

X

X

[Poveda et al., 2015]

X

X

X

X

[Poveda et al., 2015]

X

X

X

Table 2. Overview of existing ontologies concerning the main multi-agent simulation concepts

The study of multi-agent simulation approaches for model adaptation has shown the advantages of

ontologies for simulation modeling. Indeed, modeling through an ontology promotes the reusability of

the simulation model as well as its interoperability. In most of the approaches presented, model

elements are linked to simulation platform components to produce the model's simulations. Among

the ontologies studied, the ontology of [Christley et al., 2004] offers the highest and most appropriate

level of multi-agent simulation model abstraction for our approach. Although these approaches

facilitate the development of simulations through an ontology that accommodates a diversity of

simulation models, they do not address the problem of adapting the simulation model to disaster

management knowledge. However, the work of [Kruchten et al., 2007] provides some clues on the

relationships between disaster management knowledge and simulation modeling. Indeed,

infrastructures and the population, generally considered the elements at risk in case of disaster, are

the main targets of the agents' actions. The next section presents the components of a multi-agent

simulation model for disaster management.

Electronic copy available at: https://ssrn.com/abstract=3995855

3.2 Multi-agent Simulation model components

A multi-agent simulation is based on a multi-agent system. An environment, objects located in the

environment, agents that are active objects, and interactions between these different components

characterize such a system [Ferber, 1997]. Objects are passive components in the environment, and

their modeling is generally a part of the environment modeling. Therefore, there are two main

components to model in the multi-agent system: the environment (including objects inside) and

agents. Moreover, simulation aims at experimenting. That is why its modeling is composed of

parameters to configure it and observed variables to assess the experiments. This section presents,

thus, an overview of (1) Agent modeling, (2) Observed variables and assessed criteria, and finally, (3)

Environment modeling.

3.2.1 Agent modeling for disaster management simulation

Agent's concept represents active entities that can interact between them, with objects, and with their

environment. In disaster management, the Agent concept represents mainly three categories: the

affected population, disaster manager who make-decision, and responders who respond to a disaster

on the ground. Among the responders, both responders (e.g., doctors, nurses, Firefighters) and

responder engines (e.g., ambulance) can be represented by an agent as in [Saoud et al., 2006].

According to Wooldridge [Wooldridge, 2002], two main types of Agents exist: reactive and cognitive.

The reactive agents have been presented by [Brooks, 1991] as an alternative to artificial intelligence.

Inspired by the biological systems, he has defined an intelligent behavior for an agent without explicit

representation of the world and without explicit abstract reasoning. An intelligent system has been

defined as an emergent property of a complex system. It means that the Agent’s behavior results from

its interaction with the environment. Reactive Agents are generally used to represent the victims

during a simulation of disaster. The stochastic behavior is the most used approach to represent the

affected populations as in Plan-C [Narzisi et al., 2007], or as in Simgenis [Saoud et al., 2006] which uses

a Markov chain. The cellular automaton is also a technique for a reactive agent. Still, it is more used to

represent the affected population's behavior in the context of evacuation [Arai et al., 2011], [Guo and

Huang, 2008]. The reactive Agent is a simple agent that allows complex systems based on simple

entities’ interaction. Reactive agents are advantageous for large-scale simulations, but the behavior

simplicity without world representation and reasoning is also a disadvantage to represent more

complex behavior as human decision-making.

Cognitive Agents are the most classical agent type, which has a goal-directed behavior. The

approaches used to create a cognitive agent are close to artificial intelligence approaches. The most

basic cognitive Agent is the deduction reasoning agent. This Agent has a knowledge base

corresponding to its view and beliefs about the environment, a function to see the environment,

bringing new information, and updating the knowledge base [Wooldridge, 2002]. Its decision-process

is based on deduction rules to infer the actions to do from the knowledge base. This action selection

is reduced to a proof problem. This logic-based approach has the advantage of having precise and

logical semantics, which promotes its long-lived. However, the method of proof problems can have a

high complexity according to the problem's type. This high complexity implies an important time-

consuming, which is a problem in a simulation with time-constrained. A long time of computation for

the decision-process is also a rational problem if the environment has significantly changed between

the time of information-making and the action execution.

Another cognitive agent is the procedural reasoning agent. This Agent has a set of plans with a goal

corresponding to a postcondition and a context corresponding to a plan's precondition. These plans

correspond to a sequence of actions. According to its environment representation and its goal, the

Electronic copy available at: https://ssrn.com/abstract=3995855

Agent selects a set of plans, whose preconditions are satisfied and allows the accomplishment of its

purpose. It determines then one of these plans to execute it. An example of this process is presented

by [Poveda et al., 2015], where the main goal is the evacuation in indoor environments: a set of

"egress" strategies has been implemented, and their selection depends on the situation context. A

famous agent model belonging to this procedural reasoning approach is the Belief-Desire-Intention

(BDI) model, developed by [Bratman, 1987]. According to its beliefs (its beliefs and view of the

environment) and its desires (what it would like to accomplish), the BDI agent chooses a top-level goal

to pursue, which determines the agent's intentions corresponding to an action plan has decided to do.

An extended version of this model has been implemented in D-AESOP [Buford et al., 2006] for

considering the situation awareness in disaster situation management. The principal limitation of this

approach resides in this finite set of plans, which does not allow flexibility in the combination of

actions. Some multi-agent systems require agents to solve problems according to their capacities of

action.

For this requirement, a practical reasoning agent is suitable rather than a procedural reasoning agent.

Practical reasoning is composed of two main steps: (i) determining a goal to achieve according to the

environment’s state, (ii) determining how to accomplish the goal in detail in a simple manner, in the

sense of a goal is a simple action and not a plan. The ambulance agent in the AROUND project [Chu et

al., 2009] uses a decision tree to decide the action to perform, and then it uses a utility function to

determine the details of this action. For example, if the selected action is "go to the hospital," the

utility function will decide which hospital the Agent must go. However, this approach is mainly used

for more complex behavior design. The general operation of this approach corresponds to (i) observe

the environment and update its beliefs, (ii) determine the available options and filter them to choose

a goal to achieve, (iii) use means-end reasoning corresponding to artificial intelligence approaches of

planning [Wooldridge, 2002] to determine how to achieve the goal, (iv) and finally execute it.

In [Praiwattana and El Rhalibi, 2016], information about the environment (agent beliefs) and the

available actions are stored in a knowledge base. The plan design process takes this knowledge base

and the agent goal as inputs of a planner using forward chaining state-space search with heuristic

function. [Wickler et al., 2006] presents an artificial intelligence planning approach to coordinate

activities in an Emergency scenario. This approach uses an ontology called INCA to build a constraint

model. A hierarchical task network planner uses this model to determine the courses of action

resolving a problem. These authors explain the BDI model, and more precisely, the process of

intentions can be extended by using their planning approach to give more flexibility in the action

combination. The main difficulty resides in the deliberative step (ii), which manages the Agent goal.

The two extremes situations for this step are (i) to do not modify the target until it is achieved that can

create a blocked situation if the goal cannot be achieved, and (ii) change this goal anytime, leading to

a no achievement of purposes due to their abandonment. It is necessary to find a good trade-off

between these two extreme situations for the strategy of deliberation.

In disaster simulations, stakeholders can be numerous. More large is the simulation, the more there

are agents, the more simple agents must be. That is why, in evacuation simulation, the most used

agents are reactive agents. On the contrary, in strategy planning simulation, agents who decide actions

and design the planning are practical agents to allow more complex cognitive behavior. In the context

of simulating global plans application, the number of agents can become very large. That is why the

simulation model must have a maximum of reactive agents and a minimum of cognitive agents to limit

the simulation's complexity. The next paragraph presents models to represent responder agents to go

deeper in the agent modeling.

Electronic copy available at: https://ssrn.com/abstract=3995855

According to the structural comparison of disaster management among different countries, the

authors of [Labba et al., 2017] have highlighted responders' hierarchical repartition in three levels:

strategic, tactical, and operational level. This is a vertical hierarchy (Top-down), with at the top the

strategic level, and at the bottom the operational level. Thus, these authors have proposed an agent-

based meta-model for response organization structures using three different types of agents: one for

each level of the hierarchy. Despite other denominations, the study of existing multi-agent systems for

disaster response highlights multi-agent models' recurrence based on three different agents

[Hooshangi and Asghar Alesheikh, 2017], [Praiwattana and El Rhalibi, 2016], [Siebra and Tate, 2003].

These three types of agents aim at organizing or planning the response at different levels of granularity.

Table 3 presents each type of agent's denomination for each level of granularity.

References/Structure

Top

Middle

Bottom

[Labba et al., 2017]

Strategic level

Tactical level Operational level

[Hooshangi and Asghar Alesheikh, 2017] Central agent

Coordinator

Rescue

[Praiwattana and El Rhalibi, 2016]

Decision-making

Agent (DMA)

Control

agent (CA)

Field agent (FA)

[Siebra and Tate, 2003]

Strategic agent

Operational

agent

Tactical agent

Table 3. Platforms comparison according to application domains related to the disaster management domain

This structure of three different agents representing the reality and having proved its usability is

suitable to represent the different levels of responders. The techniques to represent them are

generally specific to a task and cannot be reused for other purposes. The agent model presented by

[Praiwattana and El Rhalibi, 2016] is composed of a knowledge base representing the possible actions

that an agent can do and their application condition. The decision about the sequence of actions to do

is made thanks to a planner. In the case of Preparedness, plans are sequences of actions, so a planner

is not required.

The models for responder agents shows three types of responders:

• Central agent, representing the highest level of coordination between the stakeholder managers,

• Manager agent, generally representing the coordinator of people on the ground, and

• Actor agent, representing responders on the ground.

These three agent types are mainly practical agents in strategy planning simulation to provide plans at

different detail levels. However, in the context of plan assessment, complex cognitive behavior is not

required since plans are existing. Among the different types of responders, the central and manager

agents decide to apply according to the situation. Therefore, these two agent types require a cognitive

behavior to choose the plan among a set, which is adapted to the situation. That is why procedural

agents are the most adapted to model central and manager agents in this context. Concerning actor

agents (responders on the ground), they learn and train to apply procedures and adapt them to the

situation. These agents must react according to orders coming from managers and environmental

situations. Therefore, reactive agents are adapted to model-actor agents and allow larger-scale

simulation than a representation through cognitive agents. Moreover, reactive agents are also the

most suitable to model the population, the victims, as the majority of the related works.

In a simulation model, other essential components are the observed variables and assessment criteria,

which express the simulation results and allow their assessment. The next subsection review these

components in works related to disaster management simulations.

Electronic copy available at: https://ssrn.com/abstract=3995855

3.2.2 Assessment criteria and observed variables in disaster management simulation

The review of observed variables and criteria for disaster management simulation has been applied on

ten approaches having different application scenarios: emergency and crisis response [Hawe et al.,

2015], [Praiwattana and El Rhalibi, 2016], [Walter et al., 2016], rescue strategy [Saoud et al., 2006],

[Takahashi, 2003], [Siddhartha et al., 2009], [Blatt et al., 2016], [Marecki et al., 2005], and evacuation

[Christensen and Sasaki, 2008], [Balasubramanian et al., 2006].

These approaches also have different goals: resource allocation [Hawe et al., 2015], assessing or

comparing strategies [Saoud et al., 2006], [Blatt et al., 2016], [Balasubramanian et al., 2006], [Marecki

et al., 2005] or built-environment [Christensen and Sasaki, 2008], planning response [Praiwattana and

El Rhalibi, 2016], [Walter et al., 2016], optimizing response [Takahashi, 2003], [Siddhartha et al., 2009].

Protection of elements at risk

Activity

performances

Resources

quantity

Civil

Protection

Goods Protection

References

IP

NbC

NbD

IB

IA

TP

SQ

HM

RM

[Saoud et al., 2006]

X

X

X

X

X

[Takahashi, 2003]

X

X

X

[Siddhartha et al., 2009]

X

X

X

X

X

[Hawe et al., 2015]

X

X

X

X

[Praiwattana and El Rhalibi, 2016]

X

X

[Walter et al., 2016]

X

X

[Blatt et al., 2016]

X

X

X

[Marecki et al., 2005]

X

X

[Balasubramanian et al., 2006]

X

X

[Christensen and Sasaki, 2008]

X

Table 4. Observed variables in disaster management simulation (IP: Impacted population, NbC: Number of Casualty, NbD:

Number of Dead, IB: Impacted building, IA: Impacted area, TP: Time performance, SQ: Success quantity, HM: Human means,

RM: Resource means)

Table 4 provides an overview of the variables observed by the previously stated approach. 70% of

these approaches observe the impact on the population (impacted population, number of casualties,

or dead), and 30% of them observe both the impact on the population and the impact on goods

(building or area). Population and goods are generally elements at risk, which are served by a plan or

a service in disaster management. Therefore, elements at risk served by the assessed plans can be

considered as observed variables during disaster management simulation.

The majority of these approaches (70%) assess the time performance of activities, 40% of approaches

assess the success of activities, and 20% assess both time and success of activities. This observation

shows that an activity's time performance is an essential criterion of efficiency in disaster

management.

Observed variables are essential components of a simulation model to assess the results of simulation

experiments. The study of observed variables in different simulations for disaster management allows

the identification of their type and their correspondence into a disaster management model. This

study, summarized in Table 4, highlights three links between observed variables and the disaster

management model:

1. a link to elements at risk addressed by the disaster management plan;

2. a relation to actions and tasks performed to achieve a disaster management plan;

3. a link to resource quantity to achieve a disaster management plan.

Electronic copy available at: https://ssrn.com/abstract=3995855

This information is essential to design the simulation model from the disaster management model. It

is used to elaborate on the transformation of the disaster management model into the simulation

model in this thesis's proposed solution. The elements at risk addressed by plans are observed

variables of the simulation, whose final quantity must be minimized. The duration and the success of

each action of a plan must be observed and recorded during the simulation to provide performance

metrics. Finally, resource quantity is generally a source of decision-making during disaster

management. Therefore, they are simulation input parameters for which the simulation aims at

optimizing them according to the minimization of elements at risk and the maximization of acting

performances.

3.2.3 Environment modeling for disaster management simulation

An environment is continuous through vector GIS files or discontinuous through a grid or as a network

through a graph. Grid representation is the most straightforward representation of an environment

[Hawe et al., 2015], where objects and agents are located through a grid coordinate. Moreover, it

facilitates the disaster situation representation, the agents’ interactions with their environment, and

environment effects on agents and objects by categorizing the grid cells. For example, in the simulation

model of [Saoud et al., 2006], there are three types of cells: obstacle, danger, and normal cells. These

different categories impact the evolution of victims’ health state and authorized agent actions

according to their location cell and the surrounding cells. However, it has the disadvantage of

restricting agent movement [Hawe et al., 2015]. Indeed, a continuous environment represented

through vector GIS files is much more precise. It allows using real maps with the real geometries of

environment components (e.g., polygon for buildings, points for agents, lines for roads). It has the

advantage of building a graph for a road network, allowing realistic movement of agents. In disaster

management simulation, realism and application to real use cases are essential. Therefore, creating an

environment based on vector GIS files is more suited to be more realistic for simulating real accidents

on real maps [Saoud et al., 2006]. That is why they are often used to create an environment for

evacuation simulation to create the graph of the road network.

Therefore, the simulation platform used for disaster management must allow using vector GIS files to

create a realistic environment. The next section presents a review of the multi-agent simulation

platform.

3.3 Multi-agent simulation platforms

According to their plans, agent-based simulation is the most suitable simulation model for representing

the different stakeholders acting and interacting to respond to disaster management. Therefore, the

chosen platform must be a multi-agent platform. The simulation’s goal in this system is to make a

scientific simulation based on experiments to conclude disaster management plans’ effectiveness. To

provide a scientific level of such simulation, the realism of the simulation plays a crucial role. Without

appropriate realism, the drawn conclusions cannot be used to support decision-making. Therefore, the

real-world aspect of geospatial data interpretation into the simulation is a primary requirement for

disaster management simulation.

The platform must have the capabilities to simulate the diversity of disaster management applications,

such as applying plans for city evacuation in the context of a flood. Such a type of application requires

two capabilities: large scale simulation and natural resources and environment. The large-scale ability

aims to allow the simulation of a vast number of agents, which is necessary for a city evacuation

scenario containing more than one million inhabitants. Natural resources and environment have

mainly a role in natural disaster management as floods or bushfires.

Electronic copy available at: https://ssrn.com/abstract=3995855

In conclusion, the chosen platform must be a multi-agent platform, ideally designed for scientific

multiagent simulation. The platform must allow the use of geospatial data and large-scale simulations.

It must allow scheduling and planning applications through the implementation of Belief-desire-

intention agents. Finally, it must allow natural resources and the environment to be taken into account.

The simulation platform's choice depends on the requirements previously stated for multi-agent

simulation of disaster management and the platform’s reliability.

The authors of [Kravari and Bassiliades, 2015] have made an up-to-date comparative review of agent

simulation platforms to support readers in choosing a platform. This review presents and compares 24

platforms according to a set of 23 universal criteria gathered into five categories: Platform properties,

Usability, Operating ability, Pragmatics, Security management. They also collect platforms according

to the application domains.

As presented previously, the prime pre-requisite of a platform for disaster management simulation is

to allow real-world simulation from geospatial data. Among the 24 platforms presented, the first

selection of seven platforms has been made according to their specialization or common use in the

real-world and GIS aspects. These platforms are AGLOBE [Šišlák et al., 2006], Cougaar [Helsinger and

Wright, 2005], Repast [Kravari and Bassiliades, 2015], CybelePro, SeSAm [Klügl, 2009], AnyLogic

[Borshchev, 2013], and GAMA [Grignard et al., 2013]. The more platform is designed for an application,

the more efficient it is efficient for this application. Thus, these seven platforms have been compared

according to the other requirements and application domains of disaster management. Table 5 shows

the comparison of these seven platforms according to six application domains related to disaster

management made according to Table 9 in [Kravari and Bassiliades, 2015].

AGLOBE Cougaar Repast CybelePro

SeSAm AnyLogic GAMA

Real-world and GIS

X

X

X

X

X

X

X

Large scale simulations

X

X

X

Scientific simulations

X

X

X

X

X

General purpose agent-

Based simulations

X

X

X

Scheduling and planning

X

X

X

X

Natural resources

and environment

X

X

X

Total:

1/6

2/6

2/6

4/6

5/6

5/6

6/6

Table 5. Platforms comparison according to application domains related to the disaster management domain

This comparison shows the GAMA platform as the most suitable platform for application domains

related to disaster management. The platforms CybelePro, SeSAm, and AnyLogic also provide

capabilities for most application domains related to disaster management.

The main advantage of the GAMA platform compared to SeSAm and AnyLogic platforms are to allow

large-scale simulations. This capability plays a crucial role in simulating evacuation plans of a big city.

The main advantage of GAMA compared to CybelePro is specialized in agent-based simulations and

applications in natural resources and the environment.

These four platforms (SeSAm, CybelePro, AnyLogic, and GAMA) are further analyzed using the

universal criteria of [Kravari and Bassiliades, 2015]. Concerning the pragmatics category, they have

excellent user support and are still in active development. Concerning the categories of operating

ability, platform properties, and security management, the platforms can be gathered into two

categories: one category gathering AnyLogic and CybelePro, and another one gathering SeSAm and

GAMA. The first group has the advantage of having a high operating ability globally and at least a good

Electronic copy available at: https://ssrn.com/abstract=3995855

platform security with fairness. In contrast, the second group has an excellent running ability and

average security without fairness. However, although the first group has better security management

and operating ability, it has a significant limit in the platform properties category. The first group is a

commercial, whereas the second group is free. Moreover, the level of operating ability is good enough

for prototyping. Finally, concerning the usability category, the first group has an average simplicity

whereas the second one is simple. In terms of usability, GAMA also has the advantage of being

compatible with ACL and FIPA standards, not the three other platforms.

Although the platforms as AnyLogic or CybelePro have higher security and operating ability, GAMA is

still the most suitable platform for plan experimentation for disaster management preparedness due

to its license and the capabilities that it provides. GAMA allows "complete modeling and simulation"

of large-scale simulations. It combines explicit multi-agent simulations with GIS data management,

multi-level modeling, and the capability to implement BDI and reactive agents. Thus it is powerful for

prototyping through its agent-oriented language GAML.

Besides, GAMA allows an efficient extension of the agents’ behavior using "skills" defined as additional

plugins. GAMA provides a library of "skills" assigned to an agent for diverse domains such as moving,

communication, graphics. Therefore, these skills need to be extended to the disaster management

domain to allow plan simulation.

3.4 Discussion

This section has highlighted approaches to provide flexibility in the simulation development by

proposing (1) a model or a meta-model most often specified through an ontology and (2) an

architecture based on a set of implemented components used by a simulation platform. Concerning

the simulation modeling, meta-models specified through an ontology have shown great flexibility to

represent a diversity of simulation models and facilitate their development. Indeed, the authors of

[Durak and Oren, 2016] claim that simulation engineering through the use of ontologies is the

evolution of the simulation domain. The use of ontologies for simulation modeling provides three main

advantages:

1. They are facilitating the simulation development. Approaches as [Poveda et al., 2015], Kruchten et

al., 2007], [Christley et al., 2004] use ontologies to provide flexibility in the simulation modeling

and allow the reusability of implemented simulation components. Ontologies are also used to

facilitate simulation development for other application cases as distributed simulation applications

[Benjamin et al., 2006] or other simulation techniques as discrete-event simulation with the

ontology DEMO [Miller et al., 2004]. Another interesting approach for simulation development

based on ontology has been proposed by authors of [McGinnis et al., 2011]. These authors use an

ontology to create a specific conceptual model for a problem in a domain. The user can create the

ontology implementation referred to as a domain-specific language (DSL) for a class of simulation

applications through the use of OMG SysML. Once the conceptual model is created through DSL

ontology, a model transformation is used to automate the translation to a computational

simulation model.

2. They are promoting and facilitating the sharing and reusing of simulation data. Ontologies in the

Semantic Web domain aim to describe resources to facilitate the research of relevant information.

The simulations are composed of several models, are based on data, and produce data. The sharing

of these elements can facilitate their reusability. The authors of [Lacy and Gerber, 2004] explain

that XML has been long used to interchange simulation data thanks to its advantage of solving

interchange problems. However, they promote an upgrade from XML to OWL, which overcomes

XML-only approaches' weaknesses, provides an explicit semantic, and allows the inference. The

creation of an ontology is also one approach to facilitate sharing of knowledge in the domain of

Electronic copy available at: https://ssrn.com/abstract=3995855

modeling and simulation as illustrated by the ontology of discrete-event simulation and modeling

called DEMO [Miller et al., 2004] and the ontology to describe the simulation systems engineering

process [Durak and Oren, 2016] according to the IEEE standard for Distributed Simulation

Engineering and Execution Process (DSEEP) [IEEE, 2011].

3. They are facilitating the integrated simulation with a real system into a unique system. The

common vocabulary facilitates the exchange of information between two systems: a real system

can base its operation on information representing through an ontology, which is enriched by the

integrated simulation. The authors of [Poveda et al., 2015] have designed an ontology-based

simulation that aims to be reused by other intelligent systems. The authors of [De La Asunción et

al., 2005a] present the SIADEX planning framework composed of an integrated simulation and a

real system. This SIADEX framework aims at planning firefighting according to the situation

representation. It uses an ontology called BACAREX to represent the information required for the

planning process. The integrated simulation aims to enrich the BACAREX ontology by representing

the fire situation's future state of the fire situation that constitutes an input parameter of the

planning system. The ontologies with the techniques from the Semantic Web also allow logical

reasoning operation called inference. Such a process based on ontology supports the operation of

a real system. In [Han et al., 2010], the authors use the integrated system to represent the fire

situation evolution through an ontology and use the inference process to determine hazards from

the situation representation.

From these advantages, the use of ontology for simulation modeling brings benefits both for the

sharing in the simulation community and for the flexibility of simulation modeling and development

required for plan assessment. Among existing ontologies studied in the next section, the ontology

proposed by [Christley et al., 2004] is the most suited for multi-agent simulation of disaster

management. However, it provides only high-level concepts that must be specified for the application

domain. Moreover, although approaches presented previously provides flexibility for the diversity of

simulation modeling, but they do not propose a method to design the conceptual simulation model

according to disaster management knowledge.

Although no studied approaches transform disaster management model into a multi-agent simulation

model, some simulation engineering approaches exist in other application domains that use ontology

in the process of model transformation from a domain ontology to an ontological simulation model.

The method presented by the authors of [Silver et al., 2007] uses the knowledge encoded in ontologies

to facilitate simulation modeling. The suggested technique establishes relationships between domain

ontologies and a modeling ontology. It then uses the relationships to instantiate a simulation model as

ontology instances. An application of this suggested technique has been presented by authors of [Silver

et al., 2009] for simulations based on discrete-event modeling techniques. These authors use

alignment and mapping information between the domain ontologies and the Discrete-event Modeling

Ontology (DeMO) to create DeMO instances. A code generator can then use these DeMO instances to

produce an executable simulation model. Such an approach based on alignment and mapping

information between a domain ontology and a simulation modeling ontology brings flexibility to adapt

simulation modeling according to a domain ontology. It is necessary to design a domain ontology of

disaster management to apply such a method adapted to disaster management model application and

multi-agent simulation modeling. Such research belongs to the domain of knowledge engineering

presented in the next section.

4 Knowledge engineering

Explicit knowledge in computer science generally relates to semantic knowledge representation often

used for intelligent systems (e.g., for object detection [Ponciano et al., 2019a], [Ponciano et al., 2019b],

Electronic copy available at: https://ssrn.com/abstract=3995855

[Karmacharya et al., 2015], for simulation [Poveda et al., 2015], [Durif, 2014], [Miller et al., 2004]) or

for information systems supporting decision-making (e.g. for disaster management [Shafiq et al.,

2012], [Babitski et al., 2011], [Han et al., 2010], [GeoPii, Integrasys, 2014], [Fan and Zlatanova, 2010],

[Beneito-Montagut et al., 2013]). Knowledge defines what to do and how to act, decide, or solve a

problem. Knowledge engineering consists of representing knowledge explicitly. According to Guarino

in [Guarino, 1995], it exists different levels of knowledge representation. Among these different levels,

the ontological level is used to represent the knowledge explicitly through its semantic. It aims at

creating a knowledge base that defines concepts with meaning understandable both by humans and

machines. The ontological level is chosen for its right balance between the humans’ linguistic level and

computers' logical level. And the ontological representation of knowledge is done through an ontology.

4.1 Knowledge Representation for disaster management

This section presents the existing ontologies for disaster management. In 2013, the authors of [Liu et

al., 2013] reviewed ontologies used during crisis management. This section presents seven ontologies,

presented by [Liu et al., 2013], and eight ontologies, more recent. Among these ontologies, there are

two main goals for the design of these ontologies. They are either designed for information exchange

to facilitate collaboration or for planning the response to support decision-making. The next

subsections present the existing ontologies according to these two categories of goals. They present

their advantages and disadvantages in terms of reusability.

4.1.1 Ontologies for information exchange to facilitate collaboration

The Emergel ontology [Casado et al., 2015] has been created for a European project called Disaster,

which aims to provide an international common knowledge structure allowing exchange information

about performed tasks between the different European Union countries Emergency. It covers domains

of disaster (e.g., fire, transport accident), time (e.g.timeslice), Resource (e.g. Vehicle, Equipment,

Communication), Role (e.g. BrigadeLeader, Squad leader), Infrastructure (e.g. building, street),

Organisation (e.g. FireFighting, Police, AmbulanceService), damage (e.g. 25PercentOutage), Task (e.g.

Extinguish, Transport, FloodProtection), and Movement. It is linked to FOAF and WAI vocabulary to

describe responders and their roles, as well as NeoGeo vocabulary2 for spatial objects. Its main

advantage is the completeness of the vocabulary to describe actions and tasks made on the ground.

Its disadvantage is the lack of high-level concepts for preparedness as plans description.

The MOAC ontology has been created for the management of a crisis vocabulary. MOAC ontology3

aims at defining vocabulary for the management of crisis and is expressed through RDF. It is focused

on describing an event (such as natural hazard, emergency) and humanitarian activities through "Who,

What, Where". The prime advantage of this ontology is the richness for the description of the event of

a disaster, their material aftermath, and the hazards which can result from a disaster because many

classes compose it for the infrastructure damage, problem of health, menace, the operation for a vital

resource. The principal disadvantage of this ontology is the lack of concepts for describing different

responders in disaster management. Moreover, it focuses on humanitarian activities.

The European Union has initiated the INSPIRE directive [Seifert, 2008] to create spatial information

infrastructure. The aim is to support Community environmental policies and policies or activities that

may impact the environment. A natural or another disaster often affects the environment; that is why

it is possible to find a vocabulary for a specific domain of disaster management in the INSPIRE directive.

The INSPIRE directive is ordered in 34 themes, whose "Human health and Safety" and "Natural risk

2 NeoGeo Vocabulary Specification: http://geovocab.org/doc/neogeo/, visited on 2020-09-22

3 Management of a Crisis (MOAC) Vocabulary Specification: http://www.observedchange.com/moac/ns/,

visited on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

zones". Initiatives exist to create an INSPIRE ontology4 such as the ontology5 created in the context of

heterogeneous geospatial data integration [Homburg et al., 2016]. INSPIRE covers damages (e.g.,

EventConsequence, environmental damage) and affected population (e.g., injured, evacuated,

isolated) from the theme safety; the domains of risk and hazard from the theme of natural risk zones;

operations and actors from the theme of utility and governmental services; the infrastructure domain

from themes as buildings, production, and industrial facilities, agricultural and aquacultural facilities,

and transport network; population repartition from the theme of population distribution and

demography; meteorology domain from the theme of meteorological geographical features. The

advantage of INSPIRE is that it offers a vocabulary for describing information related to disaster

preparedness such as infrastructure, governmental service, risk, population, and disaster description

such as damage and meteorology. However, INSPIRE has a lack of vocabulary to describe resources

and disaster management activities. INSPIRE is not specific to disaster management, but it is accurate

in the environment. That is why it allows gathering much information but not the representation of

disaster management.

In [Beneito-Montagut et al., 2013], the platform web “Disaster 2.0” allows actors to deposit or research

information, make requests of needs, and get answers for the application of requirements. The

information and requests added on this platform are stored in an ontology. This ontology is called Dires

and describes seven main domains: damage (an element which has been affected by a disaster),

disasters (e.g., technological, natural, conflict), geo-location (e.g., location of the incident command

post, heliport, stagging area), operations to respond to a disaster, organizations (e.g., police, fire

brigade, Business entity), responder roles (e.g., chief, commander, fireman, policeman, ambulance

man), and resources (e.g., food, clothes, vehicle, power generator). Dires is an ontology specific to the

disaster response. This ontology's advantage is the definition of the response domain through

Ressources, Operation, Casualty, Actor, Disaster, and Infrastructure. The lack of this ontology appears

at the level of risk, hazard, and plan description.

The DoRES ontology [Burel et al., 2017] aims to represent information sources, reports, and the events

and situations that occur in emergency crises. It covers domains such as events, situations,

geolocations, documents, reports, tasks, roles, organizations, and actors. This ontology gathers

concepts from ontologies SIOC, FOAF, Geoname, WGS84, and Dublin Core.

SIOC6 (Semantically-Interlinked Online Communities) describes online community sites' information as

their structure and contents to find related information and new connections between content items

and other community objects.

FOAF7 (Friend of a friend) Core describes characteristics of people and social groups independent of

time and technology. In addition to FOAF Core terms, there are terms to describe Social Web as

internet accounts, addressbooks, and other Web-based activities.

GeoNames8 ontology allows the addition of geospatial semantic information to the Word Wide Web.

It provides URI to represent a large number of geographic names and locations.

4 INSPIRE ontology from European Commission: https://inspire.ec.europa.eu/glossary/Ontology,

visited on 2020-09-22

5 INSPIRE ontology from SemGIS: https://github.com/i3mainz/SemGISOntologies, visited on 2020-09-22

6 SIOC Core Ontology Specification: https://www.w3.org/Submission/sioc-spec/, visited on 2020-09-22

7 FOAF Vocabulary Specification (0.99): http://xmlns.com/foaf/spec/, visited on 2020-09-22

8 GeoNames Ontology: http://www.geonames.org/ontology/documentation.html, visited on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

WGS849 is a vocabulary for representing latitude, longitude, and other information about spatially-

located things, using WGS84 as a reference datum.

Dublin Core10 is a vocabulary used to describe digital resources (e.g., video, images, web pages) and

physical resources (e.g., books, CD). This information is related to content (e.g., title, subject, source,

description, etc.), intellectual property (e.g., creator, contributor, editor, etc.), and instantiation (e.g.,

date, type, format, etc.).

Although the DoRES ontology describes some concepts of disaster management, there is a lack of

vocabulary to describe disaster preparedness due to their goal: the management of information

sources.

The EDXL-RESCUER ontology [Barros et al., 2015b, Barros et al., 2015a] has been developed to

coordinate and exchange information between rescuers and is based on EDXL (Emergency Data

Exchange Language), developed by the Organization for the Advancement of Structured Information

Standards (OASIS11) [Jones, 2013]. EDXL provides a set of XML-based messaging standards to improve

information sharing during an emergency like a natural disaster, for example. EDXL is based on the

National Information Exchange Model (NIEM), which is an XML-based information exchange

framework from the United States. It exists different types of message for the diverse needs: EDXL-RM

(resource message), EDXL-DE (Distribution Element), EDXL-SitRep (Situation Reporting), EDXL-TEP

(Tracking of Emergency Patients), EDXL-CAP (Common Alerting Protocol). The ontology part

corresponding to EDXL-CAP covers two main descriptions: message description (e.g., Message type,

Alert, Info) and information about the incident (e.g., category of event, resource, response type). This

ontology is limited to the description of an incident.

The EPISECC ontology [Pan and Space, 2016] has been developed in OWL to improve information

sharing through a Common Information Space. It is a Spatio-temporal ontology for modeling a common

operational picture for the first responders. It addresses the interoperability during the response. It

covers five main domains: disaster (e.g., earthquake, urban flood, flash flood), process (e.g., resource

management, physical response as decontamination, search for people), resource (e.g., financial,

human, institutional), organization (e.g., Governmental, Non-governmental, Private), common

operational picture, whose static and dynamic data (e.g.situational data: weather forecast, affected

people; operational data: resource, process). It references to GeoSPARQL, W3C Time and DOLCE-Lite

ontologies.

GeoSPARQL12 is a Geographic Query Language for RDF Data. It is an extension of SPARQL for processing

geospatial data, which "supports representing and querying geospatial data on the semantic web. [It]

defines a vocabulary for representing geospatial data in RDF. Also, GeoSPARQL is designed to

accommodate systems based on qualitative spatial reasoning and systems based on quantitative

spatial computations." [Perry and Herring, 2012]. GeoSPARQL has the same use as SPARQL, but the

difference is when it recovers a triple in a database, its declaration, serialization has not done by

RDF/XML but rather than RDF/GML. GML is a Geo Markup Language, which allows describing

geospatial data.

9 WGS84 Vocabulary: https://www.w3.org/2003/01/geo/, visited on 2020-09-22

10 Dublin Core Specification: https://www.dublincore.org/specifications/dublin-core/, visited on 2020-09-22

11 OASIS: https://www.oasis-open.org/, visited on 2020-09-22

12 GeoSPARQL (OGC Standard): http://www.ogc.org/standards/geosparql, visited on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

W3C Time13 provides the vocabulary to describe topological temporal relations, temporal reference

system (e.g. clock, calendar) time position, and duration.

DOLCE (Descriptive Ontology for Linguistic and Cognitive Engineering) [Gangemi et al., 2002] is a

foundational ontology. This ontology has a clear cognitive bias because it aims to capture the

ontological categories underlying natural language and human common sense. It is an ontology of

particulars14, divided into two categories: the enduring and perduring entities. Endurance is

continuants, which are wholly present at any time at which they exist and can change in time (e.g.,

physical objects). Perdurants or occurrents are extended in time and only partially present at any time

they exist (e.g., events and processes). DOLCE-Lite15 is a lite version of DOLCE to simplify translations

of DOLCE into various logical language.

The Humanitarian eXchange Language (HXL) ontology [Keßler and Hendrix, 2015] covers the domains

of disaster (e.g., Incident), geography (e.g., Administrative Unit), damage (e.g., affected population),

organization, and humanitarian response (e.g., displaced population). HXL is linked to vocabularies

from FOAF and GeoSPARQL. It allows mainly for describing the population impacted by an incident

(e.g., death, injured, displaced). It is limited to humanitarian activities.

Discussion on ontologies for information exchange The ontologies for information exchange to

facilitate collaboration have the main advantage of providing an extensive vocabulary to describe a

situation of crisis and response activities. However, they are generally application ontology, which

limits them to: (i) a part of response activities as humanitarian (e.g., MOAC, HXL) activities; (ii) a

situation and activities description related to information content (e.g., EDXL-RESCUER, DoRES). The

ontologies covering larger the domain (e.g., Emergel, INSPIRE, Dires, and EPISECC) do not allow the

description of Preparedness’ elements related to stakeholders’ organizational structure and the plans

to gather a set of tasks. The next section presents ontologies for planning response and aims at

searching ontologies able to represent preparedness components.

4.1.2 Ontologies for planning response

AktiveSa [Smart et al., 2007a], [Smart et al., 2007b] aims at representing humanitarian and disaster

relief operations for military agencies. It has been used by the system UICDS [Shafiq et al., 2012] to

store information for planning and rule-based reasoning to organize planning elements according to

the constraints. Even though AktiveSA was designed for military agencies, it is based on humanitarian

and disaster relief operations. In this disaster context, military agencies need to exchange information

with humanitarian agencies about: the security situation in the operations area, the locations of

humanitarian staff and facilities, the activities planned by humanitarian actors, mine action activities,

significant movements of civilians, activities of relief planned by military agencies, strike locations and

explosive munitions used during military campaigns, communication infrastructure as the best location

for radio repeaters. This model's knowledge areas are Geography, Meteorology, Activity, Humanitarian

aid, Military, Equipment, Organizations, Weapons. This model is expressed through OWL. The

advantage of AktiveSA is a high level of vocabulary description by representing a vast diversity for each

knowledge area. However, it does not provide a transparent model with defined relationships between

the different concepts.

The E-response ontology aims at describing an emergency and the response to that emergency. This

ontology is derived from the AktiveSa ontology and contains some concepts derived from DOLCE

13 W3C Time Ontology specification: https://www.w3.org/TR/owl-time/, visited on 2020-09-22

14 Particulars are entities which have no instances [Gangemi et al., 2002]

15 DOLCE-Lite Ontology: http://www.ontologydesignpatterns.org/ont/dul/DLP_397.owl, visited

on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

ontology (e.g.Endurant, Perdurant). E-Response being based on AktiveSa covers similar areas of

knowledge. Therefore, it has a similar advantage and limit. It is used to create and use a virtual

organization to respond to highly dynamic events, such as emergencies in the project Aktive

response16.

The The I-N-C-A (Issues – Nodes – Constraints –Annotations) ontology [Tate, 2003] is a constraint

model used as a shared representation of intentions for emergency response in rescue simulations

[Wickler et al., 2006] or to represent rescue actions with constrained in I-RESCUE simulation [Siebra

and Tate, 2003]. INCA provides a shared representation of the agent’s intention to coordinate activities

in an emergency response scenario. The I-N-C-A ontology is also used as a base for an intelligent

command and control system in the FIREGRID system presented in [Rein et al., 2007], to represent

tasks with constraints. This ontology is interesting to solve problems planning in emergency response

to support decision-making but is too limited to describe the preparedness components.

The IsyCri ontology [Benaben et al., 2008], [Lauras et al., 2015] is a meta-model for crisis management

represented through OWL-DL. It has been designed to gather information and knowledge about crisis

management into a crisis response coordination system. This ontology aims at characterizing a crisis

to coordinate the response process between the heterogeneous partners. It represents concepts

through three categories:

• Crisis characterization through concepts as Crisis, Factors, Effect, and Trigger;

• Studied system through concepts as Risk, Danger, Event, and Study System Components (i.e.,

Good, Civilian society, People, and Natural site);

• Treatment system through concepts as Actor, Procedure, Resource, Service, Task, Collaborative

process.

The meta-model Isycri has the advantage of a high-level description to cover the whole set of required

concepts. However, a meta-model can only be a base for further development and specification of use

cases. Thus, a meta-model provides a base for describing a diversity of use cases and for generic

reasoning on high concepts to apply knowledge on a crisis situation.

The SIADEX framework is decision support in forest fire fighting [De La Asunción et al., 2005a]. It has a

monitoring algorithm, which (1) tracks in real-time changes according to the execution of the current

plan, (2) updates the ontology, and (3) checks if the execution is like the prediction. This check aims to

detect a problem as failure execution or unexpected delay to re-plan the solution according to the

problem’s circumstance. The ontology used by this algorithm is called BACAREX. The BACAREX

ontology is an ontology of planning objects and activities related to the forest fire fighting plan.

Objects are called resources and are represented in two main classes: Material and Human. Among

materials, there are facilities and vehicles. Facilities have a GIS point as a fixed position, whereas cars

and humans have a current position due to their dynamic of the move. Information stored for every

object is operational (usage of coordinates by reasoning process of planning) and informational

(information required for by the technical staff during an episode). The BACAREX ontology contains a

fire scenario concept that forecasts weather and has a physical deployment in GIS locations specific to

fire and a concept of shifts with duration and linked to Human. This ontology is also composed of

constraints to check the consistency dynamically and detect inconsistency early. These constraints are

temporal constraints, mainly related to actions and necessary for planning, or restrictions on operating

procedures to specify resource use conditions.

16 e.Response: http://www.aiai.ed.ac.uk/project/ix/e-response/index2.html, visited on 2020-09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

The EMPATHI ontology [Gaur et al., 2019] has been created for emergency management and planning

about hazard Crisis. Super-concepts of this ontology can be gathered into three categories of

description:

• Data and information management through concepts as Modality of data, Report and Surveillance

information;

• Situation description through concepts as Place, Event, Impact, Age Group, Hazard type and phase;

• Response Activity through concepts as Involved actors, Service, Status, Facility.

EMPATHI ontology integrates vocabulary from Friend Of A Friend (FOAF)17, GeoNames18, Linked Open

Descriptions of Events (LODE) [Shaw et al., 2009], Simple Knowledge Organization System (SKOS)19,

Semantically-Interlinked Online Communities (SIOC)20, Federal Emergency Management Agency

(FEMA)21, Emergency Disasters Database (EM-DAT)22, MA-Ont23, iContact24.

The LODE is an ontology for publishing historical events as LinkedIn and mapping between other event-

related vocabularies and ontologies.

SKOS is a W3C recommendation to represent knowledge organization systems using the Resource

Description Framework (RDF). It provides a standard data model for sharing and linking knowledge

organization systems via the Web. It captures the similarity of structure and applications of knowledge

organization systems and makes it explicit to enable data and technology sharing across diverse

applications.

According to [Gaur et al., 2019], the FEMA provides a glossary of terms related to disaster preparation

and management [Anderson, 1999].

EM-DAT is a database gathering disaster events. According to [Gaur et al., 2019], it provides precise

definitions of concepts and categorizes disturbance-related events [Jonkman, 2005].

MA-Ont is a W3C Recommendation that describes a core vocabulary of properties and a set of

mappings between different metadata formats of media resources published on the Web. These

mappings aim to provide metadata representations that describe media resources' characteristics and

behavior in an interoperable manner and facilitate the sharing and reusing of metadata. iContact is an

ontology that provides basic classes and more specific properties for representing international street

addresses, phone numbers, and emails. Its benefit compared to other ontologies as FOAF is that it

considers details of international addresses, phone numbers, and emails. The EMPATHI ontology has

the advantage of providing a diversity of vocabulary to describe disaster situation and response

activities, but not to describe plans and organizational structure elaborated during the preparedness.

The Ontology-based Representation of Crisis Management Procedures for Climate Events presented

by [Kontopoulos et al., 2018] has been developed for decision support systems for crisis management.

17 FOAF Vocabulary Specification 0.99: http://xmlns.com/foaf/spec/, visited on 2020-09-22

18 GeoNames Ontology: http://www.geonames.org/ontology/documentation.html, visited on 2020-09-

22

19 Simple Knowledge Organization System (SKOS): https://www.w3.org/2004/02/skos/, visited on 2020-

09-22

20 SIOC Core Ontology Specification: http://rdfs.org/sioc/spec/, visited on 2020-09-22

21 Federal Emergency Management Agency (FEMA): https://www.fema.gov/, visited on 2020-09-22

22 Emergency Disasters Database (EM-DAT): https://www.emdat.be/index.php, visited on 2020-09-22

23 Ontology for Media Resources 1.0: https://www.w3.org/ns/ma-ont, visited on 2020-09-22

24 International Contact Ontology: http://ontology.eil.utoronto.ca/icontact.html, visited on 2020-

09-22

Electronic copy available at: https://ssrn.com/abstract=3995855

It covers the representation of a crisis, climate parameters that may cause climate crises, sensor

analysis, first responder unit allocations, crisis incidents, and related impacts. This ontology is limited

to disasters related to climate.

The ontologies for planning address the response step, and it provides mainly a large vocabulary to

describe response activities. However, as their goal is to results in planning, they do not give vocabulary

to define plans related to response activities. Only IsyCri ontology has concepts related to plans as a

procedure and provides a global view of prepared response activities. The particularity of IsyCri is to

be a meta-model. A meta-model has the advantage of being composed of super-concepts linked

between them by super-properties. Super-concepts and super-properties allow them to gather a vast

diversity of specifications. Such a high level of description provides a base to define different disaster

management models. In addition to IsyCri ontology, it also exists some other meta-models for disaster

management as a global disaster management meta-model presented in [Othman and Beydoun, 2013]

and a meta-model for each phase of disaster management presented by [Othman et al., 2014] and

used by [Othman and Beydoun, 2016]. Among these different meta-models, the meta-model of

[Othman and Beydoun, 2013] provides the most adapted concepts related to preparedness

components and their application.

4.1.3 Discussion

Reusable ontologies have been identified from analyzing the main required terms and the study of

these ontologies. Table 6 and Table 7 summarizes the analysis of existing ontologies according to the

main terms required. On the one hand, the meta-model presented by [Othman and Beydoun, 2013]

has been identified as the most suitable for the main terms representation at a high level of

description. On the other hand, the Emergel, Dires and EPISECC ontologies are very complete in terms

of concepts for the description of the elements contained in a plan, such as the tasks and resources

that can intervene in applying a plan. Among these three ontologies, Emergel ontology is the most

interesting to complement the high-level concepts of [Othman and Beydoun, 2013] because of its

design related to tactical symbols. This advantage facilitates the extraction of knowledge from tactical

plans corresponding to maps containing tactical symbols. Thus, the use of the high-level concepts of

[Othman and Beydoun, 2013] allows the representation of a wide variety of plans based on a wide

variety of tasks, roles, resources whose concepts are described by the Emergel ontology. The

knowledge base using these ontologies can then be fulfilled by knowledge extracted from data.

Therefore, the next section presents the existing methods for knowledge extraction and integration.

Table 6. Existing ontologies according to required terms (1)

Electronic copy available at: https://ssrn.com/abstract=3995855

Table 7. Existing ontologies according to required terms (2)

4.2 Knowledge extraction and integration

The conventional approach for integrating data into a system (e.g., data warehouse, information

system, knowledge base) is the Extract-Transform-Load (ETL) approach. This approach has been

promoted in the 1970s for managing data warehouses. It is often used nowadays in the semantic

domain, as shown by works of [Bansal and Kagemann, 2015] in the context of big data integration. The

extract step consists of extracting the raw values of the data. The transformation consists of structuring

data into a targeted schema. The transformation step also cleans data by removing redundant or

useless information, groups information, and checks information. Finally, the load step consists in the

insertion of data into the targeted system.

In the context of heterogeneous data integration into an ontology, each data source is transformed

into a local ontology gathering the extracted raw values. The local ontologies can either be linked to a

global ontology and thus be directly loaded into the global ontology or require a new transformation

based on a mapping step with the global ontology to be then loaded [Cruz and Xiao, 2005], [Hacherouf

et al., 2015]. This step of transformation by ontology mapping is also a process used to integrate

knowledge and information represented into another ontology model.

Such a process is also used in the Semantic Web to enrich linked data, as shown by works of [Debruyne

et al., 2017]. This work presents a methodology of enriching linked data with a geospatial dimension

from CSV files [Repici, 2006]. The first process of this methodology that they call uplift transforms the

geospatial data into RDF triples. The second one enriches data and consists of link discovery and their

incorporation into RDF data. In addition to the ETL approach principle, this methodology proposes the

last step to produce new datasets. This last step is called downlift and transforms the newly enriched

RDF data into an enriched CSV file.

Many works in the integration of heterogeneous data concern geospatial data due to the diversity of

data formats and the heterogeneity of information that they can contain. Geospatial data are also the

primary data containing information related to disaster management preparedness. Thus, this section

presents mainly approaches related to geospatial data integration.

In the domain of semantic and geospatial integration, the GeoKnow project [Grange et al., 2014], which

aims at geographically enriching data with linked data web, gathers several tools to apply the

methodology presented by [Debruyne et al., 2017]. This project uses TripleGeo [Patroumpas et al.,

2014] and Sparqlify [Stadler et al., 2013] as tools to do the uplift process. The enrichment is done using

Electronic copy available at: https://ssrn.com/abstract=3995855

LIMES [Ngomo and Auer, 2011] for link discovery and Geolift for enriching and data cleaning. The end-

user of GeoKnow aims at managing linked data on the web, so no downlift step is included. Instead,

the authors visualized the data in a user interface.

These examples from the Semantic Web illustrate the two main steps of integration. These steps are

the uplift process to transform data content into RDF triples and the ontology mapping process used

for link discovery in the Semantic Web, for transforming a local ontology into a global ontology in the

ETL process, and for knowledge integration from another ontology model.

4.2.1 Uplift process

The most common uplift processes are semi-automatic approaches based on schema matching to

transform data into RDF triples directly loadable into the global ontology. Such semi-automatic

approaches are used in projects as Karma [Knoblock et al., 2012], which can process heterogeneous

data and Silk [Volz et al., 2009] specialized for data reading of RDF, CSV, and XML files to convert data

into the RDF form. However, the schema matching approaches are generally specific to a data type.

Semi-automatic approaches for database Many techniques and standards are created for information

integration from a database to an ontology. The W3C has developed the R2RML standard, a language

to express relational databases into RDF triples [Das et al., 2012]. An RDF graph represents RDF

mapping. Thus, R2RML expresses a customized mapping from relational databases to RDF data sets.

R2RML can also be used to express a mapping from a CSV file as presented in the paper [Debruyne et

al., 2017]. Another similar approach is DB2OWL, which is presented in [Ghawi and Cullot, 2007], which

aims to generate an ontology mapping from a relational database. Some tools like BOOTOX [Jimènez-

Ruiz et al., 2015] have been developed to facilitate the mapping from given relational databases to

extract a corresponding ontology from the database schema. The tool Sparqlify included in the

GeoKnow project [Grange et al., 2014] provides an RDF view through a SPARQL query, using SPARQL

to SQL translation mechanisms. Similarly, [Rodríguez-Muro and Rezk, 2015] presents an approach to

access the data in a database. This approach uses R2RML, not to convert or translate the database's

content in an ontology but to translate a SPARQL request (used to request an ontology) into a SQL

request. Concerning approaches used in disaster management system, SOKNOS in [Paulheim et al.,

2009] presents a semi-automatic approach, which is a conversion of an interactive mapping by the

user between data from database and resource ontology into F-logic rules. The COBACORE data

framework [GeoPii, Integrasys, 2014] comprises a service provider that uses a domain ontology to

transform a relational database into RDF triples.

Semi-automatic approaches for geospatial data Several approaches have been developed to convert

the content of geospatial data into an RDF model. Some approaches are specialized for one type of

storage, like the approach GML2RDF presented in [Casado et al., 2015], which is specialized for

translating GML format. [Bizid et al., 2014] presents another approach for GML files. This approach

converts GML data sets to local ontologies using GML schemas and provides automated interlinking

strategies for similarly structured database resources.

Automatic approaches are also available for automatic uplift. Some approaches are specialized for

relational databases as the direct mapping presented by the W3C [Arenas et al., 2011]. Some others

are able to process heterogeneous formats (e.g. databases, CSV [Repici, 2006], GML [Burggraf, 2006],

shapefile [Environmental Systems Research Institute (ESRI), 1998]) as the Datalift project [Scharffe et

al., 2012]. The uplift process of this project converts the input format into RDF triples (subject-

predicate-object). The subject corresponds to an element of a row. The predicate is based on the

column name. The object is the cell's content corresponding to the intersection between the row of

the subject and the predicate column. Then, it converts RDF triples into a "well-formed RDF" according

Electronic copy available at: https://ssrn.com/abstract=3995855

to chosen vocabularies by using SPARQL CONSTRUCT queries. [Pinkel et al., 2017] presents a schema

matching based on intermediate graphs obtained by transforming the two inputs corresponding to a

relational database and an ontology. These two intermediate graphs are then matched. Their approach

uses two matchers based on the graph structure and a lexical one. First, it creates a matching using a

pairwise connectivity graph to gather pair by pair of potential nodes. It then applies a Jaccard, similarity

matcher [Niwattanakul et al., 2013] and finally, applies a structure matcher using an adaptation of

similarity flooding algorithms [Melnik et al., 2002]. Other automatic schema matching approaches are

presented in the survey [Rahm and Bernstein, 2001] and [Rahm, 2011].

4.2.2 Ontology mapping process

The authors of [Choi et al., 2006] make the distinction between the three following ontology

mapping:

• Ontology mapping between an integrated global ontology and local ontologies "is used to map a

concept found in one ontology into a view, or a query over other ontologies (e.g., over the global

ontology in the local-centric approach, or over the local ontologies in the global-centric

approach)."

• Ontology mapping between local ontologies "is the process that transforms the source ontology

entities into the target ontology entities based on semantic relation. The source and target are

semantically related at a conceptual level."

• Ontology mapping in ontology merge and alignment "establishes correspondence among source

(local) ontologies to be merged or aligned, and determines the set of overlapping concepts,

synonyms, or unique concepts to those sources. This mapping identifies similarities and conflicts

between the various source (local) ontologies to be merged or aligned."

The first type of ontology mapping (between an integrated global ontology and local ontologies) can

be used to access local ontologies based on the knowledge base's vocabulary and resulting from the

uplift process. However, these local ontologies are generally directly integrated into the knowledge

base through, for example SPARQL Update query.

The second type of ontology mapping (between local ontologies) requires a semantic definition of

relationships between concepts of the source and target ontologies. This semantic definition can be

expressed in OWL or through semantic rules.

The third type of ontology mapping (in ontology merge and alignment) requires to establish

correspondences between ontologies. The process to establish correspondences between ontologies

is called ontology matching and aims at solving link discovery problems.

Different techniques of ontology matching are reviewed in [Otero-Cerdeira et al., 2015]. Their

classification is divided between the element-level and structure-level, but also between semantic and

syntactic techniques. Another method of classification is to use the kind of input rather than the

granularity interpretation. In this case, the classification of techniques is divided between context-

based techniques, semantic or syntactic, and content-based techniques, terminological, structural,

extensional, or semantic.

[Nentwig et al., 2017] surveyed current link discovery frameworks and discussed eleven frameworks

by highlighting their specificities. All of the presented frameworks support the relation owl:sameAs,

but only the Silk Framework [Volz et al., 2009] and LIMES [Hillner and Ngomo, 2011] allow the user to

specify other relations. The majority of them require manual configuration. However, four of them

have a semi-automatic and adaptive linking specification based on the data set analysis and the

identification of the most discriminative properties. Among the four learning-based frameworks, three

Electronic copy available at: https://ssrn.com/abstract=3995855

use supervised learning, whereas only two utilize unsupervised learning. Concerning similarity

measures, the eleven frameworks all utilize them, but only five have a structure matcher. Two of them

are interesting when it comes to the geospatial domain: Zhishi.links [Niu et al., 2011] and LIMES

(GeoKnow project) because they use geographical coordinates as a similarity measure. These similarity

measures intervene in the primary step of link discovery, which is the ontology matching.

As shown by the study of the link discovery frameworks, the matching techniques are generally

combined to obtain better results. The work of [Do and Rahm, 2002] aims at comparing the efficiency

of the different types of matchers and assesses their combinations. Their benchmark highlights the

efficiency of a combination of matchers and matcher results' reusability to simplify future mapping.

This third type of ontology mapping aims at merging ontologies, as the merge of local ontologies into

a global ontology to constitute the knowledge base or at aligning ontologies to access various

ontologies as it is done in the architecture federated of [Farias et al., 2015] which uses SWRL-Rule

selection for ontology interoperability.

4.2.3 Discussion

A knowledge base can be fulfilled by the integration of knowledge extracted from data. An integration

process is composed of uplift and ontology mapping processes. An end-to-end plan assessment system

requires an automatic process of uplift to process heterogeneous data without requiring supervision.

This requirement aims to integrate knowledge extracted from all stakeholders’ data without requiring,

for example, a new mapping specific to the database of stakeholders. Among the automatic uplift

processes, most of them are specialized in one type of data [Arenas et al., 2011]. Only a few approaches

allow the processing of heterogeneous data. Datalift, an automatic approach for heterogeneous data,

has been tested to assess its usability for the proposed solution. However, the RDF graph resulting

from its uplift contains only annotation properties; it has not produced an RDF graph composed of

individuals linked by properties. This quality of uplift is not enough to extract disaster management

knowledge. Therefore, there is a lack of the automatic integration approaches of knowledge extracted

from heterogeneous data.

5 Conclusion

Disaster management consists of four steps: mitigation, preparedness, response, and recovery. Among

these steps, the response step is the most critical since human lives depend on this phase’s

effectiveness. The response effectiveness depends on plan effectiveness defined in upstream during

the preparedness step. Therefore, the preparedness step and the assessment of plan effectiveness

during the preparedness phase are important research areas to avoid problems during the response

step. The plan effectiveness assessment required (1) the identification of their application conditions,

(2) some metrics to quantify their effectiveness, and (3) their experimentation through exercises or

computer simulation.

The identification of plan application conditions required to cluster situations for which a plan has

similar effectiveness. The common features between such situations form the conditions impacting

the plan’s effectiveness. In the case of a large-scale plan application or extensive testing of such plans,

the clustering and analysis of characteristics should be unsupervised to avoid human analysis prone to

error when the number of situations and characteristics is significant.

[Bayram et al., 2012] and [Larsson, 2008] define different metrics to assess the plans’ effectiveness

according to the four main areas defined by the Sphere Project [The Sphere Project, 2011] (health,

housing, food and nutrition, water and sanitation). However, these metrics depend on the plan’s

purpose and should be defined within the plan’s scope. It means that they must be observed during

Electronic copy available at: https://ssrn.com/abstract=3995855

the plan’s experimentation phase. The different situations characteristics of plan experimentation

must be put in correlation. A common characteristics analysis can achieve such correlation to

determine the characteristics that impact the plans. The plan’s experimentation phase is carried out

through exercises or computer simulations. Computer simulations have an advantage over exercises,

allowing a high number of experiments to assess plans. Multi-agent simulations are the most suitable

for plan experiments [Mishra et al., 2019]. However, most multi-agent simulation approaches for

disaster management [Christensen and Sasaki, 2008], [D’Orazio et al., 2014], [Zhou et al., 2012], [Mas

et al., 2015], [Nagarajan et al., 2012], [Mishra et al., 2019], [Hawe et al., 2012], [Saoud et al., 2006],

[Poveda et al., 2015] are limited. Indeed, their case-dependent simulation modeling produces a lack of

adaptability from one situation to another. Yet, the testing of disaster management plans requires the

adaptation of simulations to address disaster management scenarios' diversity and complexity.

Therefore some approaches [Poveda et al., 2015], [Kruchten et al., 2007], [Christley et al., 2004] allow

simulation adaptation to a diversity of action strategies. Such approaches use ontologies to model the

simulation. These ontologies represent the concepts of the simulation domain at a high level of

abstraction in order to allow the modeling of a wide variety of strategies. They are used to automate

simulation development and have the advantage of facilitating interoperability with other systems.

However, these approaches do not allow simulations to be modeled based on disaster management

knowledge. Nevertheless, ontologies' use to represent the simulation modeling and the disaster

management knowledge allows taking advantage of Semantic Web technologies and ontologies.

Indeed, the Semantic Web technologies can allow defining mapping and alignment between the two

ontologies to design simulation modeling instances according to disaster management instances.

The study of disaster management ontologies shows the benefits of using high-level concepts of the

meta-model presented by [Othman et al., 2014] to allow the definition of a wide variety of plans, but

also its need for specification. On the other hand, it shows the advantages of the ontology Emergel

[Casado et al., 2015] for its comprehensiveness in describing concepts at a low level. These ontologies

provide the terminology to describe disaster management knowledge. However, the knowledge is

represented through assertions. Some parts of knowledge are stored through data. The study of

existing knowledge extraction approaches from data has highlighted an interesting project, which is

Datalift [Scharffe et al., 2012]. However, Datalift creates annotation assertions, which is not the most

adapted RDF representation to integrate knowledge.

This simulation modeling process requires the explicit representation of disaster management

knowledge and a high-level representation of simulation modeling concepts capable of

accommodating a wide variety of simulation models. The study of approaches to facilitate simulation

adaptation has identified the ontology presented by [Christley et al., 2004] as the most relevant

ontology for simulation modeling. However, this ontology must be completed to specify the high-level

simulation modeling concepts according to the modeling needs for disaster management simulation.

Such simulations must be composed of at least three responder agents’ type (central, manager, and

actor), GIS environment with information such as the demography, the critical infrastructure, area of

risk. They should observe the 12 types of variables defined by [Bayram et al., 2012] for which plan is

assessed according to their purpose.

Among different multi-agent simulation platforms (AGLOBE, Cougaar, Repast, CybelePro, SeSAm, Any-

Logic, and GAMA) the GAMA platform allows real-word and GIS representation, large scale

simulations, scientific simulation, general-purpose agent-based simulation scheduling and planning,

natural resources and environment and thus, appears to be the most suitable platform for the

simulation of plans compared with the other studied approaches. The GAMA platform does not have

an agent’s behavior set specific to the disaster management domain. Therefore, it is necessary to

Electronic copy available at: https://ssrn.com/abstract=3995855

extend the agent behaviors by various functionalities typical of the disaster management domain. Such

an extension can be carried out by the addition of an external plugin call "skills".

6 References

[Anderson, 1999] Anderson, C. V. (1999). The Federal Emergency Management Agency. Disaster Prevention and

Management: An International Journal, 8(1):499–509.

[Arai et al., 2011] Arai, K., Harsono, T., and Basuki, A. (2011). Cellular Automata for Traffic Modelling and Simulations in a

Situation of Evacuation from Disaster Areas. Cellular Automata - Simplicity Behind Complexity, page 193.

[Arenas et al., 2011] Arenas, M., Bertails, A., Prud’hommeaux, E., and Sequeda, J. F. E. (2011). A Direct Mapping of Relational

Data to RDF. W3C Working Draft 20 September 2011, 27 (September 2012):1–11.

[Bayram et al., 2012] Bayram, J. D., Kysia, R., and Kirsch, T. D. (2012). Disaster Metrics: A Proposed Quantitative Assessment

Tool in Complex Humanitarian Emergencies - The Public Health Impact Severity Scale (PHISS). PLoS Currents.

[Babitski et al., 2011] Babitski, G., Bergweiler, S., Grebner, O., Oberle, D., Paulheim, H., and Probst, F. (2011). SoKNOS - Using

semantic technologies in disaster management software. Lecture Notes in Computer Science (including subseries Lecture

Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 6643 LNCS (PART 2):183–197.

[Balasubramanian et al., 2006] Balasubramanian, V., Massaguer, D., Mehrotra, S., and Venkata-subramanian, N. (2006).

DrillSim: A simulation framework for emergency response drills. In Lecture Notes in Computer Science (including subseries

Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), volume 3975 LNCS, pages 237–248. Springer.

[Barros et al., 2015b] Barros, R., Kislansky, P., Salvador, L., Almeida, R., Breyer, M., Pedraza, L. G., and Vieira, V. (2015b). EDXL-

RESCUER ontology: An update based on faceted taxonomy approach. CEUR Workshop Proceedings, 1442. [BBK, 2015] BBK

(2015). Grundlagen Krisenmanagement. Technical report, Bundesamt für Bevölkerungsschutz und Katastrophenhilfe.

[Benaben et al., 2008] Benaben, F., Hanachi, C., Lauras, M., Couget, P., and Chapurlat, V. (2008). A metamodel and its ontology

to guide crisis characterization and its collaborative management. In Proceedings of ISCRAM 2008 - 5th International

Conference on Information Systems for Crisis Response and Management, pages 189–196.

[Beneito-Montagut et al., 2013] Beneito-Montagut, R., Shaw, D., and Brewster, C. (2013). Web 2.0 and Social Media in

Disaster Management: Using Web 2.0 applications and Semantic Technologies to strengthen public resilience to disasters.

UK: Aston University, page 174.

[Benjamin et al., 2006] Benjamin, P., Patki, M., and Mayer, R. (2006). Using ontologies for simulation modeling. In Proceedings

- Winter Simulation Conference, pages 1151–1159. IEEE.

[Bansal and Kagemann, 2015] Bansal, S. K. and Kagemann, S. (2015). Integrating Big Data: A Semantic Extract-Transform-Load

Framework. Computer, 48(3):42–50.

[Bharosa et al., 2010] Bharosa, N., Lee, J., and Janssen, M. (2010). Challenges and obstacles in sharing and coordinating

information during multi-agency disaster response: Propositions from field exercises. Information Systems Frontiers,

12(1):49–65.

[Bizid et al., 2014] Bizid, I., Faiz, S., Boursier, P., and Yusuf, J. C. M. (2014). Integration of heterogeneous spatial databases for

disaster management. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and

Lecture Notes in Bioinformatics), volume 8697, chapter Integratio, pages 77–86. Springer International Publishing, Cham.

[Blatt et al., 2016] Blatt, F., Becker, M., and Szczerbicka, H. (2016). Analysing the cost-efficiency of the multi-agent flood

algorithm in search and rescue scenarios. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial

Intelligence and Lecture Notes in Bioinformatics), volume 9872 LNAI, pages 147–154. Springer.

[Borshchev, 2013] Borshchev, A. (2013). The big book of simulation modeling: multimethod modeling with AnyLogic 6.

AnyLogic North America.

[Boufedji et al., 2018] Boufedji, D., Guessoum, Z., Brandão, A., Ziadi, T., and Mokhtari, A. (2018). Towards a MAS Product Line

Engineering Approach. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and

Lecture Notes in Bioinformatics), volume 10738 LNAI, pages 161–179. Springer.

Electronic copy available at: https://ssrn.com/abstract=3995855

[Buford et al., 2006] Buford, J., Jakobson, G., Lewis, L., Parameswaran, N., and Ray, P. (2006). D-AESOP: A Situation-AWARE

BDI Agent System for Disaster Situation Management. In Proceedings In AAMAS Workshop on Agent Technology for Disaster

Management, Japan, pages 1–23.

[Burggraf, 2006] Burggraf, D. S. (2006). Geography markup language. Data Science Journal, 5:178–204.

[Burel et al., 2017] Burel, G., Piccolo, L. S., Meesters, K., and Alani, H. (2017). DoRES-A three-tier ontology for modelling crises

in the digital age. In Proceedings of the International ISCRAM Conference, volume 2017-May, pages 834–845.

[Casado et al., 2015] Casado, R., Rubiera, E., Sacristan, M., Schütte, F., and Peters, R. (2015). Data interoperability software

solution for emergency reaction in the Europe Union. Natural Hazards and Earth System Sciences, 15(7):1563–1576.

[Chaffey and White, 2010] Chaffey, D. and White, G. (2010). Business Information Management: Improving Performance

Using Information Systems. FT Prentice Hall, Harlow.

[Choi et al., 2006] Choi, N., Song, I.-Y., and Han, H. (2006). A survey on ontology mapping. ACM Sigmod Record, 35(3):34–41.

[Christensen and Sasaki, 2008] Christensen, K. and Sasaki, Y. (2008). Agent-Based emergency evacuation simulation with

individuals with disabilities in the population. Jasss, 11(3):9.

[Christley et al., 2004] Christley, S., Xiang, X., and Madey, G. (2004). An ontology for agent-based modeling and simulation. In

Proceedings of the agent 2004 conference.

[Chu et al., 2009] Chu, T. Q., Drogoul, A., Boucher, A., and Zucker, J. D. (2009). Interactive learning of independent experts’

criteria for rescue simulations. Journal of Universal Computer Science, 15(13):2719–2743.

[Coppola, 2011] Coppola, D. P. (2011). Introduction to International Disaster Management. Elsevier.

[Cruz and Xiao, 2005] Cruz, I. F. and Xiao, H. (2005). The role of ontologies in data integration. Engineering Intelligent Systems,

13(4):245–252.

[Das et al., 2012] Das, S., Sundara, S., and Cyganiak, R. (2012). R2RML: RDB to RDF Mapping Language. W3C Recommendation,

pages 1–34.

[Debruyne et al., 2017] Debruyne, C., McGlinn, K., McNerney, L., and O’Sullivan, D. (2017). A lightweight approach to explore,

enrich and use data with a geospatial dimension with semantic web technologies. In GeoRich 2017 - 4th International ACM

Workshop on Managing and Mining Enriched Geo-Spatial Data, in Conjunction with SIGMOD 2017, pages 1–6. ACM.

[De La Asunción et al., 2005a] De La Asunción, M., Castillo, L., Fdez-Olivares, J., García-Pérez, O., González, A., and Palao, F.

(2005a). SIADEX: An interactive knowledge-based planner for decision support in forest fire fighting. AI Communications,

18(4):257–268.

[Do and Rahm, 2002] Do, H.-H. and Rahm, E. (2002). COMA — A system for flexible combination of schema matching

approaches. In VLDB ’02: Proceedings of the 28th International Conference on Very Large Databases, pages 610–621. VLDB

Endowment.

[D’Orazio et al., 2014] D’Orazio, M., Quagliarini, E., Bernardini, G., and Spalazzi, L. (2014). EPES - Earthquake pedestrians’

evacuation simulator: A tool for predicting earthquake pedestrians’ evacuation in urban outdoor scenarios. International

Journal of Disaster Risk Reduction, 10(PA):153–177.

[Dunn, 1973] Dunn, J. C. (1973). A fuzzy relative of the isodata process and its use in detecting compact well-separated

clusters. Journal of Cybernetics, 3(3):32–57.

[Durak and Oren, 2016] Durak, U. and Oren, T. (2016). Towards an ontology for simulation systems engineering. Simulation

Series, 48(2):92–99.

[Durif, 2014] Durif, T. (2014). Environnement informé sémantiquement enrichi pour la simulation multi-agents: application à

la simulation en environnement virtuel 3D. PhD thesis, Dijon.

[Environmental Systems Research Institute (ESRI), 1998] Environmental Systems Research Institute (ESRI) (1998). ESRI

Shapefile Technical Description. Computational Statistics, 16(July):370–371.

[Fan and Zlatanova, 2010] Fan, Z. and Zlatanova, S. (2010). Exploring ontology potential in emergency management. In GI4DM

2010 Conference - Geomatics for Crisis Management, pages 132–137. Politechno di Torino.

Electronic copy available at: https://ssrn.com/abstract=3995855

[Farias et al., 2015] Farias, T. M., Roxin, A., and Nicolle, C. (2015). FOWLA, A federated architecture for ontologies. In Lecture

Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics),

volume 9202, pages 97–111. Springer.

[Federal Emergency Management Agency, 2010] Federal Emergency Management Agency (2010). Developing and

maintaining emergency operations plans. Comprehensive Preparedness Guide 101.

[Ferber, 1997] Ferber, J. (1997). Les systèmes multi-agents: vers une intelligence collective. InterEditions.

[Gangemi et al., 2002] Gangemi, A., Guarino, N., Masolo, C., Oltramari, A., and Schneider, L. (2002). Sweetening ontologies

with DOLCE. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture

Notes in Bioinformatics), volume 2473, pages 166–181. Springer.

[Gaur et al., 2019] Gaur, M., Shekarpour, S., Gyrard, A., and Sheth, A. (2019). Empathi: An Ontology for Emergency Managing

and Planning about Hazard Crisis. In Proceedings - 13th IEEE International Conference on Semantic Computing, ICSC 2019,

pages 396–403. IEEE.

[GeoPii, Integrasys, 2014] GeoPii, Integrasys, T. (2014). D2.2 Definition of Modelling Hierarchy. Technical report, COBACORE

- European Commission.

[Ghawi and Cullot, 2007] Ghawi, R. and Cullot, N. (2007). Database-to-Ontology Mapping Generation for Semantic

Interoperability. In VDBL’07 conference, VLDB Endowment ACM, volume 91, pages 1—-8.

[Grange et al., 2014] Grange, J. J. L., Lehmann, J., Athanasiou, S., Rojas, A. G., Giannopoulos, G., Hladky, D., Isele, R., Ngonga

Ngomo, A.-C., Sherif, M. A., Stadler, C., and Wauer, M. (2014). The GeoKnow Generator: Managing Geospatial Data in the

Linked Data Web. Proceedings of the Linking Geospatial Data Workshop.

[Grignard et al., 2013] Grignard, A., Taillandier, P., Gaudou, B., Vo, D. A., Huynh, N. Q., and Drogoul, A. (2013). Gama 1.6:

Advancing the art of complex agent-based modeling and simulation. In International conference on principles and practice of

multi-agent systems, pages 117–131. Springer.

[Guarino, 1995] Guarino, N. (1995). Formal ontology, conceptual analysis and knowledge representation. International

Journal of Human - Computer Studies, 43(5-6):625–640.

[Guha et al., 2000] Guha, S., Rastogi, R., and Shim, K. (2000). Rock: A robust clustering algorithm for categorical attributes.

Information systems, 25(5):345–366.

[Guha et al., 2001] Guha, S., Rastogi, R., and Shim, K. (2001). Cure: an efficient clustering algorithm for large databases.

Information Systems, 26(1):35 – 58.

[Guha-Sapir and Below, 2002] Guha-Sapir, D. and Below, R. (2002). The quality and accuracy of disaster data: A comparative

analyse of 3 global data sets. Disaster Management facility, World Bank, Working paper ID.

[Gulati et al., 2012] Gulati, R., Wohlgezogen, F., and Zhelyazkov, P. (2012). The Two Facets of Collaboration: Cooperation and

Coordination in Strategic Alliances. Academy of Management Annals, 6(1):531–583.

[Guo and Huang, 2008] Guo, R. Y. and Huang, H. J. (2008). A modified floor field cellular automata model for pedestrian

evacuation simulation. Journal of Physics A: Mathematical and Theoretical, 41(38):385104.

[Hacherouf et al., 2015] Hacherouf, M., Bahloul, S. N., and Cruz, C. (2015). Transforming XML documents to OWL ontologies:

A survey. Journal of Information Science, 41(2):242–259.

[Han et al., 2010] Han, L., Potter, S., Beckett, G., Pringle, G., Welch, S., Koo, S. H., Wickler, G., Usmani, A., Torero, J. L., and

Tate, A. (2010). FireGrid: An e-infrastructure for next-generation emergency response support. Journal of Parallel and

Distributed Computing, 70(11):1128–1141.

[Hawe et al., 2015] Hawe, G. I., Coates, G., Wilson, D. T., and Crouch, R. S. (2015). Agent-based simulation of emergency

response to plan the allocation of resources for a hypothetical two-site major incident. Engineering Applications of Artificial

Intelligence, 46:336–345.

[Helsinger and Wright, 2005] Helsinger, A. and Wright, T. (2005). Cougaar: A robust configurable multi agent platform. In 2005

IEEE Aerospace Conference, pages 1–10. IEEE.

[Hillner and Ngomo, 2011] Hillner, S. and Ngomo, A. C. N. (2011). Parallelizing LIMES for large-scale link discovery. In ACM

International Conference Proceeding Series, pages 9–16. ACM.

Electronic copy available at: https://ssrn.com/abstract=3995855

[Hooshangi and Asghar Alesheikh, 2017] Hooshangi, N. and Asghar Alesheikh, A. (2017). Agent-based task allocation under

uncertainties in disaster environments: An approach to interval uncertainty. International Journal of Disaster Risk Reduction,

24:160–171.

[Homburg et al., 2016] Homburg, T., Prudhomme, C., Würriehausen, F., Karmacharya, A., Boochs, F., Roxin, A., and Cruz, C.

(2016). Interpreting heterogeneous geospatial data using semantic web technologies. In Computational Science and Its

Applications – ICCSA 2016, pages 240–255. Springer International Publishing.

[Jimènez-Ruiz et al., 2015] Jimènez-Ruiz, E., Kharlamov, E., Zheleznyakov, D., Horrocks, I., Pinkel, C., Skjæveland, M. G.,

Thorstensen, E., and Mora, J. (2015). BOOTOX: Practical mapping of RDBs to OWL 2. In Lecture Notes in Computer Science

(including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), volume 9367, pages 113–

132. Springer.

[Jones, 2013] Jones, E. (2013). Emergency data exchange language (edxl) distribution element version 2.0 - committee

specification 02.

[Jonkman, 2005] Jonkman, S. N. (2005). Global perspectives on loss of human life caused by floods. Natural Hazards,

34(2):151–175.

[Karmacharya et al., 2015] Karmacharya, A., Boochs, F., and Tietz, B. (2015). Knowledge guided object detection and

identification in 3D point clouds. In Videometrics, Range Imaging, and Applications XIII, volume 9528, page 952804.

International Society for Optics and Photonics.

[Karypis et al., 1999] Karypis, G., Eui-Hong Han, and Kumar, V. (1999). Chameleon: hierarchical clustering using dynamic

modeling. Computer, 32(8):68–75.

[Kontopoulos et al., 2018] Kontopoulos, E., Moßgraber, J., Van Der Schaaf, H., Lombardo, F., Ferri, M., Vrochidis, S., Mitzias,

P., Hertweck, P., Hilbring, D., Norbiato, D., Karakostas, A., and Kompatsiaris, I. (2018). Ontology-based representation of crisis

management procedures for climate events. In Proceedings of the International ISCRAM Conference, volume 2018-May,

pages 1051–1062.

[Kravari and Bassiliades, 2015] Kravari, K. and Bassiliades, N. (2015). A survey of agent platforms. Jasss, 18(1):11.

[Kruchten et al., 2007] Kruchten, P., Woo, C., Monu, K., and Sotoodeh, M. (2007). A human-centered conceptual model of

disasters affecting critical infrastructures. In Intelligent Human Computer Systems for Crisis Response and Management,

ISCRAM 2007 Academic Proceedings Papers, pages 327–344.

[Klügl, 2009] Klügl, F. (2009). Sesam: Visual programming and participatory simulation for agent-based models.

[Knoblock et al., 2012] Knoblock, C. A., Szekely, P., Ambite, J. L., Goel, A., Gupta, S., Lerman, K., Muslea, M., Taheriyan, M.,

and Mallick, P. (2012). Semi-automatically mapping structured sources into the semantic web. In Lecture Notes in Computer

Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), volume 7295 LNCS,

pages 375–390. Springer.

[Labba et al., 2017] Labba, C., Salah, N. B., and Bellamine Ben Saoud, N. (2017). An agent-based meta-model for response

organization structures. In Lecture Notes in Business Information Processing, volume 301, pages 154–167. Springer.

[Larsson, 2008] Larsson, A. (2008). A Framework for Evaluating Emergency Preparedness Plans and Response Strategies. IIASA

Interim Report.

[Lacy and Gerber, 2004] Lacy, L. and Gerber, W. (2004). Potential modeling and simulation applications of the Web Ontology

Language-owl. In Proceedings - Winter Simulation Conference, volume 1, pages 265–270. IEEE.

[Lauras et al., 2015] Lauras, M., Truptil, S., and Bénaben, F. (2015). Towards a better management of complex emergencies

through crisis management meta-modelling. Disasters, 39(4):687–714.

[Liu et al., 2013] Liu, S., Brewster, C., and Shaw, D. (2013). Ontologies for crisis management: A review of state of the art in

ontology design and usability. In ISCRAM 2013 Conference Proceedings - 10th International Conference on Information

Systems for Crisis Response and Management, pages 349–359.

[Malone and Crowston, 1990] Malone, T. W. and Crowston, K. (1990). What is coordination theory and how can it help design

cooperative work systems? In Proceedings of the 1990 ACM Conference on Computer-Supported Cooperative Work, CSCW

1990, pages 357–370. ACM.

Electronic copy available at: https://ssrn.com/abstract=3995855

[McGinnis et al., 2011] McGinnis, L., Huang, E., Kwon, K. S., and Ustun, V. (2011). Ontologies and simulation: A practical

approach. Journal of Simulation, 5(3):190–201.

[MacQueen et al., 1967] MacQueen, J. et al. (1967). Some methods for classification and analysis of multivariate observations.

In Proceedings of the fifth Berkeley symposium on mathematical statistics and probability, pages 281–297. Oakland, CA, USA.

[Marecki et al., 2005] Marecki, J., Schurr, N., and Tambe, M. (2005). Agent-Based Simulations for Disaster Rescue Using the

DEFACTO Coordination System, chapter 14, pages 281–297. John Wiley & Sons, Ltd.

[Mas et al., 2015] Mas, E., Koshimura, S., Imamura, F., Suppasri, A., Muhari, A., and Adriano, B. (2015). Recent Advances in

Agent-Based Tsunami Evacuation Simulations: Case Studies in Indonesia, Thailand, Japan and Peru. Pure and Applied

Geophysics, 172(12):3409–3424.

[Melnik et al., 2002] Melnik, S., Garcia-Molina, H., and Rahm, E. (2002). Similarity flooding: A versatile graph matching

algorithm and its application to schema matching. In Proceedings - International Conference on Data Engineering, pages 117–

128. IEEE.

[Miller et al., 2004] Miller, J. A., Baramidze, G. T., Sheth, A. P., and Fishwick, P. A. (2004). Investigating ontologies for simulation

modeling. In Proceedings of the IEEE Annual Simulation Symposium, pages 55–63. IEEE Computer Society.

[Mishra et al., 2019] Mishra, D., Kumar, S., and Hassini, E. (2019). Current trends in disaster management simulation modelling

research. Annals of Operations Research, 283(1-2):1387–1411.

[Nagarajan et al., 2012] Nagarajan, M., Shaw, D., and Albores, P. (2012). Disseminating a warning message to evacuate: A

simulation study of the behaviour of neighbours. European Journal of Operational Research, 220(3):810–819.

[Nance, 1994] Nance, R. E. (1994). The Conical Methodology and the evolution of simulation model development. Annals of

Operations Research, 53(1):1–45.

[Narzisi et al., 2007] Narzisi, G., Mincer, J. S., Smith, S., and Mishra, B. (2007). Resilience in the face of disaster: Accounting

for varying disaster magnitudes, resource topologies, and (sub)population distributions in the PLAN C emergency planning

tool. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in

Bioinformatics), volume 4659 LNAI, pages 433–446. Springer.

[Nentwig et al., 2017] Nentwig, M., Hartung, M., Ngonga Ngomo, A. C., and Rahm, E. (2017). A survey of current Link Discovery

frameworks. Semantic Web, 8(3):419–436.

[Niu et al., 2011] Niu, X., Rong, S., Zhang, Y., and Wang, H. (2011). Zhishi.links results for OAEI 2011. CEUR Workshop

Proceedings, 814:220–227.

[Niwattanakul et al., 2013] Niwattanakul, S., Singthongchai, J., Naenudorn, E., and Wanapu, S. (2013). Using of jaccard

coefficient for keywords similarity. Lecture Notes in Engineering and Computer Science, 2202(6):380–384.

[Ng and Han, 2002] Ng, R. T. and Han, J. (2002). Clarans: A method for clustering objects for spatial data mining. IEEE

transactions on knowledge and data engineering, 14(5):1003–1016.

[Ngomo and Auer, 2011] Ngomo, A. C. N. and Auer, S. (2011). LIMES - A time-efficient approach for large-scale link discovery

on the web of data. In IJCAI International Joint Conference on Artificial Intelligence, pages 2312–2317.

[Otero-Cerdeira et al., 2015] Otero-Cerdeira, L., Rodríguez-Martínez, F. J., and Gómez-Rodríguez, A. (2015). Ontology

matching: A literature review. Expert Systems with Applications, 42(2):949 – 971.

[Othman and Beydoun, 2013] Othman, S. H. and Beydoun, G. (2013). Model-driven disaster management. Information &

Management, 50(5):218–228.

[Othman and Beydoun, 2016] Othman, S. H. and Beydoun, G. (2016). A metamodel-based knowledge sharing system for

disaster management. Expert Systems with Applications, 63:49–65.

[Othman et al., 2014] Othman, S. H., Beydoun, G., and Sugumaran, V. (2014). Development and validation of a Disaster

Management Metamodel (DMM). Information Processing and Management, 50(2):235–271.

[Pan and Space, 2016] Pan, E. and Space, E. I. (2016). D4 . 4 . – Ontology model for the EPISECC use case. Technical report,

EPISECC project, grant agreement [607078].

Electronic copy available at: https://ssrn.com/abstract=3995855

[Patroumpas et al., 2014] Patroumpas, K., Alexakis, M., Giannopoulos, G., and Athanasiou, S. (2014). TripleGeo: An ETL tool

for transforming geospatial data into RDF triples. In CEUR Workshop Proceedings, volume 1133, pages 275–278.

[Paulheim et al., 2009] Paulheim, H., Döweling, S., Tso-Sutter, K., Probst, F., and Ziegert, T. (2009). Improving usability of

integrated emergency response systems: the SoKNOS approach. INFORMATIK 2009 - Im Focus das Leben, Beitrage der 39.

Jahrestagung der Gesellschaft fur Informatik e.V. (GI), 154:1435–1449.

[Perry and Herring, 2012] Perry, M. and Herring, J. (2012). OGC GeoSPARQL-A geographic query language for RDF data.

[Pinkel et al., 2017] Pinkel, C., Binnig, C., Jimenez-Ruiz, E., Kharlamov, E., Nikolov, A., Schwarte, A., Heupel, C., and Kraska, T.

(2017). INCMap: A Journey towards ontology-based data integration. Lecture Notes in Informatics (LNI), Proceedings - Series

of the Gesellschaft fur Informatik (GI), 265:145–164.

[Praiwattana and El Rhalibi, 2016] Praiwattana, P. and El Rhalibi, A. (2016). Survey: Development and analysis of a games-

based crisis scenario generation system. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial

Intelligence and Lecture Notes in Bioinformatics), volume 9654, pages 85–100. Springer.

[Ponciano et al., 2019a] Ponciano, J.-J., Karmacharya, A., Wefers, S., Atorf, P., and Boochs, F. (2019a). Connected Semantic

Concepts as a Base for Optimal Recording and Computer-Based Modelling of Cultural Heritage Objects, volume 18. Springer

International Publishing.

[Ponciano et al., 2019b] Ponciano, J.-J., Trémeau, A., and Boochs, F. (2019b). Automatic detection of objects in 3d point clouds

based on exclusively semantic guided processes. ISPRS International Journal of Geo-Information, 8(10).

[Poveda et al., 2015] Poveda, G., Serrano, E., and Garijo, M. (2015). Designing emergency management services by ontology

driven social simulation. IT CoNvergence PRActice (INPRA), 3(1):17–32.

[Rahm, 2011] Rahm, E. (2011). Towards large-scale schema and ontology matching. In Schema matching and mapping, pages

3–27. Springer.

[Rahm and Bernstein, 2001] Rahm, E. and Bernstein, P. A. (2001). A survey of approaches to automatic schema matching.

VLDB Journal, 10(4):334–350.

[Repici, 2006] Repici, D. J. (2006). The comma separated value (csv) file format. Creativyst Software.

[Rein et al., 2007] Rein, G., Abecassis Empis, C., and Carvel, R. (2007). The Dalmarnock Fire Tests: Experiments and Modelling.

School of Engineering and Electronics, University of Edinburgh.

[Rodrigues Da Silva, 2015] Rodrigues Da Silva, A. (2015). Model-driven engineering: A survey supported by the unified

conceptual model. Computer Languages, Systems and Structures, 43:139–155.

[Rodríguez-Muro and Rezk, 2015] Rodríguez-Muro, M. and Rezk, M. (2015). Efficient SPARQL-to-SQL with R2RML mappings.

Journal of Web Semantics, 33:141–169.

[Saoud et al., 2006] Saoud, N. B.-B., Mena, T. B., Dugdale, J., Pavard, B., and Ahmed, M. B. (2006). Assessing large scale

emergency rescue plans: an agent based approach. The International Journal of Intelligent Control and Systems, 11(4):260–

271.

[Saxena et al., 2017] Saxena, A., Prasad, M., Gupta, A., Bharill, N., Patel, O. P., Tiwari, A., Er, M. J., Ding, W., and Lin, C.-T.

(2017). A review of clustering techniques and developments. Neurocomputing, 267:664 – 681.

[Scharffe et al., 2012] Scharffe, F., Atemezing, G., Troncy, R., Gandon, F., Villata, S., Bucher, B., Hamdi, F., Bihanic, L., Képéklian,

G., Cotton, F., Euzenat, J., Fan, Z., Vandenbussche, P. Y., and Vatant, B. (2012). Enabling linked data publication with the

datalift platform. In AAAI Workshop - Technical Report, volume WS-12-13, pages 25–30.

[Seifert, 2008] Seifert, M. (2008). INfrastructure for SPatial InfoRmation in Europe INSPIRE.

[Shafiq et al., 2012] Shafiq, B., Chun, S. A., Atluri, V., Vaidya, J., and Nabi, G. (2012). Resource sharing using UICDS™ framework

for incident management. Transforming Government: People, Process and Policy, 6(1):41–61.

[Shaw et al., 2009] Shaw, R., Troncy, R., and Hardman, L. (2009). Lode: Linking open descriptions of events. In Gómez-Pérez,

A., Yu, Y., and Ding, Y., editors, The Semantic Web, pages 153–167, Berlin, Heidelberg. Springer Berlin Heidelberg.

[Siddhartha et al., 2009] Siddhartha, H., Sarika, R., and Karlapalem, K. (2009). Score Vector : A New Evaluation Scheme for

RoboCup Rescue Simuation Competition 2009. Evaluation.

Electronic copy available at: https://ssrn.com/abstract=3995855

[Siebra and Tate, 2003] Siebra, C. and Tate, A. (2003). I-Rescue: A Coalition Based System to Support Disaster Relief

Operations. Proceedings of the Third International Conference on Artificial Intelligence and Applications.

[Silver et al., 2007] Silver, G. A., Al-Haj Hassan, O., and Miller, J. A. (2007). From domain ontologies to modeling ontologies to

executable simulation models. In Proceedings - Winter Simulation Conference, pages 1108–1117. IEEE Press.

[Šišlák et al., 2006] Šišlák, D., Rehák, M., Pechoucek, M., and Pavlícek, D. (2006). Deployment of a globe multi-agent platform.

In Proceedings of the fifth international joint conference on Autonomous agents and multiagent systems, pages 1447–1448.

[Smart et al., 2007a] Smart, P. R., Russell, A., and Shadbolt, N. R. (2007a). AKTiveSA: Supporting civil-military information

integration in Military Operations Other Than War. In 2007 International Conference on Integration of Knowledge Intensive

Multi-Agent Systems, KIMAS 2007, pages 434–439. IEEE.

[Smart et al., 2007b] Smart, P. R., Russell, A., Shadbolt, N. R., Shraefel, M. C., and Carr, L. A. (2007b). AKTiveSA: A technical

demonstrator system for enhanced situation awareness. Computer Journal, 50(6):703–716.

[Stadler et al., 2013] Stadler, C., Unbehauen, J., Lehmann, J., and Auer, S. (2013). Connecting Crowd-sourced Spatial

Information to the Data Web with Sparqlify.pdf. Technical report, University of Leipzig.

[Tate, 2003] Tate, A. (2003). <{INCA}>: an ontology for mixed-initiative synthesis tasks. In Proceedings of the workshop on

mixed-initiative intelligent systems ({MIIS}) at the international joint conference on artificial intelligence ({IJCAI-03}),

Acapulco, Mexico, pages 125–130.

[Takahashi, 2003] Takahashi, T. (2003). RoboCupRescue simulation league. In Lecture Notes in Artificial Intelligence (Subseries

of Lecture Notes in Computer Science), volume 2752, pages 477–481. Springer.

[The Sphere Project, 2011] The Sphere Project (2011). Humanitarian Charter and Minimum Standards in Humanitarian

Response. Practical Action.

[Volz et al., 2009] Volz, J., Bizer, C., Gaedke, M., and Kobilarov, G. (2009). Silk - A Link Discovery Framework for the Web of

Data. CEUR Workshop Proceedings, 538.

[Walter et al., 2016] Walter, H., Gering, F., Arnold, K., Gerich, B., Heinrich, G., and Welte, U. (2016). Rodos-based simulation

of potential accident scenarios for emergency response management in the vicinity of nuclear power plants. BfS-SCHR-60/16,

pages 0221–2016091214084.

[Wickler et al., 2006] Wickler, G., Tate, A., and Potter, S. (2006). Using the< INCA> Constraint Model as a Shared

Representation of Intentions for Emergency Response. In Proceedings of the First International Workshop on Agent

Technology for Disaster Management (ATDM), at the Fifth International Joint Conference on Autonomous Agents and

Multiagent Systems (AAMAS 2006), Future University, Hakodate, Japan, pages 2–9. Fifth International Joint Conference on

Autonomous Agents and Multiagent Systems.

[Wooldridge, 2002] Wooldridge, M. (2002). An introduction to multiagent systems. John Wiley & Sons.

[Zhang et al., 1996] Zhang, T., Ramakrishnan, R., and Livny, M. (1996). Birch: an efficient data clustering method for very large

databases. ACM sigmod record, 25(2):103–114.

[Zhou et al., 2012] Zhou, T., Wang, D., Zong, D., and He, X. (2012). The Study of the responsibility space regionalization of

emergency shelters in the urban center and the evaluation of evacuation road based on microscope computer simulation - A

case of Chongqing three gorges square area. Disaster Advances, 5(4):230–236.

Electronic copy available at: https://ssrn.com/abstract=3995855