Driving Towards a Sustainable Future: A Multi-Layered Agent-Based

Digital Twin Approach for Rural Areas

Stephanie C. Rodermund

, Annegret Janzso

, Ye Eun Bae

, Anna Kravets

, Alexander Schewerda

Jan Ole Berndt

and Ingo J. Timm

1,2

Cognitive Social Simulation, German Research Center for Artiﬁcial Intelligence, Behringstr. 21, 54296 Trier, Germany

Business Informatics, Trier University, Universitaetsring 15, 54296 Trier, Germany

Keywords:

Digital Twin, Agent Decision-Making, Sustainability, Rural Development.

Abstract:

The production of CO

, a major contributor to global emissions, is signiﬁcantly caused by human activities,

with transportation accounting for approximately 25% of worldwide emissions. Fostering pro-environmental

behaviors (PEB) is vital for achieving emission reduction. Individual decision-making to adopt PEBs is com-

plex, inﬂuenced by personal characteristics, situational factors, and externally sourced information. This posi-

tion paper introduces a conceptual framework for a multi-layered agent-based Digital Twin (DT) designed to

facilitate experimentation with various scenarios and intervention approaches promoting PEB among residents

in rural regions. As a use case, we outline how to apply the DT to a speciﬁc rural area in Germany.

1 INTRODUCTION

Climate change has been a signiﬁcant concern over

the past decade and escalating global temperatures are

precipitating adverse consequences such as extreme

weather and natural disasters (Shukla et al., 2019).

Carbon dioxide (CO

), among various greenhouse

gases, stands out as one of the predominant contrib-

utors to climate change (National Academies of Sci-

ences et al., 2020). Numerous human activities con-

tribute substantially to the production of CO

, with

the transportation sector alone responsible for approx-

imately one-quarter of global CO

emissions (Agency

and Environment, 2022).

Under the 2015 Paris Agreement, nations agreed

to cap the rise in global average temperature at 1.5

◦

above pre-industrial levels, thereby preventing more

pronounced and deleterious consequences of climate

change (Masson-Delmotte et al., 2022). To bring

about a meaningful reduction in emissions within a

limited timeframe, it becomes imperative not only to

substitute fossil fuels with other alternative fuels in

the long-term but also to change human behavior in

the short-term (Chapman, 2007).

An international survey reveals that 85% of

German respondents believe that climate change is

mostly caused by human activities (Leiserowitz et al.,

2022). This awareness, however, does not always

translate into pro-environmental actions. For in-

stance, while the environmental beneﬁts of public

transport over private transport are known, many still

opt for the latter, preferring immediate convenience

to long-term sustainability. Such social dilemma,

choosing between immediate and long-term reward

(Dawes, 1980), is particularly evident in rural areas of

Germany, where private transport is often favored due

to infrastructural inadequacy (S

uddeutsche Zeitung,

2022).

We aim to provide a framework that enables ex-

perimenting with various scenarios and intervention

approaches promoting pro-environmental behaviors

(PEB) among residents in rural areas, with a focus

on transportation. This approach is grounded in the

capabilities of Digital Twin (DT) to support decision-

making in promoting sustainable behaviors with a vir-

tual representation of the area and implementing sim-

ulation as well as reasoning (cf. Wang et al. 2023).

The DT framework is designed to offer insight into

how individuals make decisions under a complex in-

terplay of their internal and external factors without

requiring a change into the actual physical infrastruc-

ture.

The paper is structured as follows: Section 2 gives

a brief introduction to DT and discusses the results of

a systematic literature review. In Section 3, the pro-

posal of our conceptual DT framework - consisting of

a spatial, individual and social layer - is introduced.

For this, we represent individuals with artiﬁcial agents

under consideration of cognitive and social concepts

using Agent-based Social Simulation (ABSS). In Sec-

tion 4, the DT framework is applied to a use case of

386

Rodermund, S., Janzso, A., Bae, Y., Kravets, A., Schewerda, A., Berndt, J. and Timm, I.

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas.

DOI: 10.5220/0012460100003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 1, pages 386-395

ISBN: 978-989-758-680-4; ISSN: 2184-433X

Table 1: Search Queries and Results from Structured Liter-

ature Research.

”Digital Twin” AND Hits Relevant

”Urban” 37

”City” 43

”Smart City” 33

”Town” 1

”Rural” 1

”Agent” 33

”Sustainability” OR

”Sustainable”

”Emission” 3

Backward 75

Forward 11

Sum 268 19

a rural area. Finally, Section 5 summarizes the paper

and addresses the limitations and future work.

2 BACKGROUND: DIGITAL

TWINS OF INHABITED AREAS

DTs were initially conceptualized as a virtual or dig-

ital equivalent to a physical product. The concept

was ﬁrst mentioned in the context of product lifecy-

cle management (Grieves, 2014). This transforma-

tive concept has found a variety of applications and

continues to evolve into new industries and use cases.

Various types of DTs exist, from device-level mod-

els for individual machines to city-level replicas for

urban planning. The concept of DT is constantly

evolving, leading to a lack of a uniﬁed consolidated

deﬁnition that distinguishes it from other technolo-

gies (VanDerHorn and Mahadevan, 2021). When DTs

are applied to urban settings, usually entire cities are

replicated, enabling the analysis and optimization of

complex systems such as transportation networks and

infrastructure.

To get an insight into current DT approaches of

inhabited areas, we conducted a structured literature

review using a simpliﬁed version of the snowballing

technique, originally introduced by Wohlin (Wohlin,

2014). We focused on the computer science bibli-

ography dblp

using several search queries in con-

junction with “Digital Twin” (cf. Table 1). Be-

yond queries with general terms synonymous with

populated regions, we incorporated “Agent” as a

speciﬁc term to appropriately represent individuals.

Agent-based Modeling (ABM) and Multi-agent sys-

tems (MAS) allow for investigating the effect of in-

dividual characteristics on decision-making on the

micro level and emergent behavior on the macro

https://dblp.org/

level which arises from communication and interac-

tion (Bonabeau, 2002). We retrieved 268 publications

in total using a backward and forward search, from

which we identiﬁed 19 as relevant. The following

subsection analyzes the publications in terms of rele-

vant components for representing rural areas and dis-

cusses our research focus.

2.1 Aspects of Digital Twins of

Inhabited Areas

As stated in Section 1, human activities have a signif-

icant impact on CO

emissions. Hence, in addition

to spatial circumstances, individual and collective be-

havior need to be addressed when providing a DT

framework that enables experimenting with aspects

impacting PEB. Therefore, we analyzed the retrieved

publications with a special focus on spatial, individ-

ual, and social aspects. Table 2 displays an overview

of the occurrence of those aspects in the publications.

Spatial Aspects. Taking a closer look at spatial as-

pects is relevant for assessing the proximity of in-

dividuals to their preferred locations such as pub-

lic transportation or supermarkets, calculating pop-

ulation density, and dividing the overall population

into groups such as households, communes, neigh-

borhoods. Some key aspects thereof are granularity,

structure, and representation within the DT.

Granularity is concerned with how the area is rep-

resented. This can be on the level of individual dis-

tricts or neighborhoods, or more ﬁne-grained on the

level of individual blocks or buildings. Major et al.

(2021) represent geographical as well as building as-

Table 2: Spatial, Individual and Social Aspects in Publica-

tions.

spatial

aspects

individual

aspects

social

aspects

Barat et al. (2022) X X X

Clemen et al. (2021) X X

Adreani et al. (2023) X

Yun et al. (2023) X X

Mohammadi and Taylor (2019) X X X

Ahn et al. (2020) X X

Mavrokapnidis et al. (2021) X X

Bujari et al. (2021) X X

Meta et al. (2021) X X

Major et al. (2022) X X

Pan et al. (2022) X

Sottet et al. (2022) X

Mohammadi and Taylor (2020) X

Van Den Berghe (2021) X

Major et al. (2021) X

Le Fur et al. (2023) X X X

Bellini et al. (2022) X

Ferreira et al. (2013) X

Fan et al. (2022) X X

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas

387

pects, while the DT of Adreani et al. (2023) consists

of individual buildings for the purpose of a photore-

alistic model of the city. Van Den Berghe (2021) and

Mohammadi and Taylor (2020) include most of the

physical objects in their DT. Le Fur et al. (2023) oper-

ates on a ﬁner level, and even includes a detailed rep-

resentation of interiors, in addition to the buildings.

A ﬁne level of granularity has also been chosen for

infrastructure representation by Major et al. (2022),

Ferreira et al. (2013) and Bellini et al. (2022), giving

information i.e. on road type, direction, and capacity,

whereas Pan et al. (2022) show a high-level focus on

infrastructure.

The structure describes how items are represented

and distributed under the chosen granularity. For ex-

ample, Barat et al. (2022) present the city as areas

with different types of buildings and numbers of peo-

ple in each area. The areas thereby differ from each

other in the ratio of building types, to represent resi-

dential or industrial areas. Clemen et al. (2021) and

Bujari et al. (2021) structure DTs by coordinates, giv-

ing the start and end point of each commute, thereby

eliminating the need of a detailed representation of

infrastructure. The structure can also be presented as

a graph with nodes at geographic coordinates, which

can then be connected via links and structured in indi-

vidual sections (Yun et al., 2023; Sottet et al., 2022).

Finally, it is important to determine what needs

to be represented within the DT. This can include in-

frastructure and points of interest (POIs), as well as

vegetation or terrain aspects. As an example, Adreani

et al. (2023) choose a very ﬁne-grained representa-

tion for their DT, and thereby included buildings, road

shapes, names, paths, areas, terrain, and other struc-

tures, as well as markers for POIs.

Individual Aspects. Addressing individual aspects

is crucial in the context of inhabited areas, especially

with regard to behavioral adoption concerning CO

reduction. This was diversely reﬂected in half of the

publications we reviewed. For instance, the inﬂuence

of individuals can be regarded on a community level

by including an aggregated impact of their behavior

on the respective city (cf. Mavrokapnidis et al. 2021;

Bujari et al. 2021). Another option is to only include

the results of an analysis concerning a speciﬁc group

of people. For example, Ahn et al. (2020) investigate

the distress of elderly people in pedestrian situations

and identify the best paths for minimal emotional or

physical distress. Mohammadi and Taylor (2019) use

a game-theoretic approach including groups of stake-

holders (citizens, government, industry) that represent

diverse interests. Fan et al. (2022) focus on the indi-

vidual level and deﬁne trajectories of people in the

city in a two-stage process using a coarse and ﬁne-

grained level to predict human mobility.

Among publications that deal with human factors

six of them use ABM. Agents here utilize rather sim-

ple, mostly reactive, architectures, as these are of-

ten used for generating crowd behavior on the macro

level. Le Fur et al. (2023) map agents’ behavior with

existing data to deﬁne daily routines of real inhabi-

tants of a city, and abstract to behavioral groups, e.g.,

moving between locations, sleeping or eating. Pa-

pers with health-related approaches focus on where

agents interact with each other to compute the prob-

ability of disease transmission. For instance, Barat

et al. (2022) model agents using characteristics such

as demographic data and profession to deﬁne behav-

ioral patterns, i.e., movements and amount or type of

contacts. Similarly, publications in transport focus

on the presence of agents at a particular location and

how they move between places, e.g., Clemen et al.

(2021) model agent decisions in favor of the speciﬁc

type of transport based on travel time required. Meta

et al. (2021) test the adaptability of a City Physiol-

ogy framework designed for urban DTs within the

testbed of a virtual stadium. Pedestrian movement

patterns are affected by an individual’s characteristics

and knowledge as well as changes in the environment,

e.g., an event or behavioral change of other agents.

Yun et al. (2023) develop a DT for testing scenarios

regarding garbage collection in a city to optimize the

operating policy. The agent decision model is a binary

decision about whether to dump trash into containers

or empty a full container.

Social Aspects. When modeling inhabited areas,

not only individual behavior but also potential inter-

actions between those individuals are relevant as they

result in emergent effects. There were four publica-

tions that utilized some kind of social relationship be-

tween individuals. Those publications consider so-

cial behavior as a result of simple contact at a cer-

tain location. Barat et al. (2022) use contacts to deter-

mine infection spread in the population. Here, agents

are given archetypes to implement different contact

patterns. Meta et al. (2021) include social behavior

by implementing crowd dynamics considering move-

ment directions. A similar approach is used by Le Fur

et al. (2023), where human-animal interaction occurs.

A detailed model for human agents is not elaborated

upon. The only paper that mentions more complex

concepts such as the values of citizens still remains on

a simple level, viewing citizens as merely consumers

of the infrastructure services (Mohammadi and Tay-

lor, 2019).

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

388

Our Research Focus. It is noticeable that the term

“rural” returned only one result in the literature

search, which indicates the need for additional re-

search. To build upon insights from papers we re-

trieved, we aim to incorporate existing research and

address some of its drawbacks. We intend to oper-

ate at a less detailed spatial level compared to the

retrieved publications. Since our focus is on trans-

portation, representing terrain is less relevant for our

framework, we rather put an emphasis on POIs and

sustainable infrastructure like public transportation.

Our DT is intended to be used as an experimental

framework, i.e., to compare scenarios and test the ef-

fect of interventions on behaviors and the overall sys-

tem. For this, agents must be capable of deliberation

considering options, which was not tackled in detail

in the retrieved publications. Furthermore, they often

overlook the dynamic nature of the intricate aspects

of social organization and structure. Incorporating in-

sights from social theorists could enhance the models’

ability to simulate and predict outcomes in real-world

scenarios more accurately, especially in the context

where human behavior and social structures play a

pivotal role (cf. Helbing 2012). Whether and how

these aspects could be implemented in a DT will be

discussed in the following section.

3 A MULTI-LAYERED DIGITAL

TWIN APPROACH FOR RURAL

AREAS

When creating a DT of an inhabited area, an enor-

mous amount of data from multiple sources at dif-

ferent granularity has to be integrated. In examin-

ing an inhabited area, we focus on what is classiﬁed

as a “complex system”, (cf. Helbing 2012). One of

the ways to approach a complex system is via struc-

tural complexity, hence a multi-layer framework. Fig-

ure 1 depicts the conceptual layers of our DT frame-

work with the spatial layer at the bottom. Addition-

ally, an essential component of a DT is the popula-

tion of the focused area. Individual behaviors (indi-

vidual layer) and relationships between people (so-

cial layer) can impact the overall system and lead to,

e.g., ﬂuctuations of CO

emissions. This may be trig-

gered by actions that have a direct impact (e.g., the

choice of means of transportation) or an indirect im-

pact (e.g., information exchange between people that

might affect their decision-making). The decision for

or against PEB does not only depend on individual

preferences, goals, and habits, but also on situational

circumstances or information gathered from external

Spatial Layer

Social Layer

Individual Layer

Attitude

Subjective Norm

Perceived

Behavioral Control

Intention Behavior

Environment

Local Network

Social Network

Strengths of

Attitudes

Lifestyles/

Actortypes

Environmental

Awareness

Cost vs. Benefit

Local

Circumstances

Personal

Circumstances

Self Efficacy

reflect decide execute

action

Figure 1: Layers of the DT Structure.

sources. These complex relationships can be realized

by MAS and ABM as it has established itself in the

representation of cognitive decision-making in vary-

ing application areas (Bonabeau, 2002; An, 2012).

This requires the use of additional psychological con-

cepts. The box in the center connects individual and

spatial layers and considers the social structure, which

emerges as a result of connections between individu-

als and their spatial and social proximity.

Layers comprising our DT are introduced in the

following order: the spatial layer in Section 3.1, the

individual layer in Section 3.2 and the connecting so-

cial layer in Section 3.3.

3.1 Spatial Layer

One main aspect of the spatial layer is the possibil-

ity to represent location, and thereby show individu-

als at given places, where they can interact with one

another. Generally, locations in this layer are repre-

sented as cities, places, and POIs. The infrastructure

network deﬁned by streets and public transport op-

tions connects individual locations. The overall pop-

ulation is distributed in residential- and workplaces.

As described in Section 2.1, this layer is presented

in different levels of granularity. Here, we take an

approach of representing multiple towns to demon-

strate commutes between places. A region is further

deﬁned by individual zones and areas (cf. Barat et al.

2022). At the ﬁner level of granularity, zones are fur-

ther divided into individual neighborhoods. Further,

a neighborhood is depicted as a group of buildings.

This is a common approach when building a DT of a

city (cf. Adreani et al. 2023; Le Fur et al. 2023), es-

pecially when they incorporate a strong visual compo-

nent. Higher levels of granularity are of advantage, as

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas

389

they enable a thorough overview of the relevant steps

while keeping processing times low. Other aspects,

such as areas of vegetation or details on terrain, are

omitted at this stage.

Our focus is on a rural region in Germany, con-

sisting of several towns and villages. This is why

it is essential to consider places of work such as of-

ﬁces and POIs. These are determined for each region

with OpenStreetMap

(OSM) data with relevant tags

of places located within the area. In our use case, this

is particularly relevant since key locations might not

be within a single town, requiring residents to travel

farther to adhere to their daily routines. Therefore,

we use data reﬂecting the positions of relevant points

in locations and infrastructure, which can be obtained

from OSM data.

3.2 Individual Layer

Similar to spatial aspects, the complexity of portray-

ing individual decisions depends on the application

context and purpose of DTs. For instance, in ur-

ban, energy, or mobility planning, the use of reactive

behavior models of individuals or groups is usually

sufﬁcient to represent the impact of small structural

changes that affect the overall system. As the focus of

our research is identifying suitable intervention strate-

gies for an inhabited area with varying spatial and

infrastructural circumstances (e.g., state of the pub-

lic transport) as well as generational diversity, a more

elaborated representation of individuals is required.

In the case of our DT framework, this implies that

the individual behavior is modeled two-fold. First,

agents are equipped with a realistic time schedule rep-

resenting their daily life including the speciﬁc geo-

graphic places where the agent is spending time, e.g.,

the workplace or POIs, as well as how the agent trav-

els to these places (e.g., by car or public transport)

(cf. Le Fur et al. 2023; Barat et al. 2022). Second,

agents have a deliberative component that refers to

PEB in the speciﬁc scenarios. These situations can

occur at any step of the daily schedule, e.g., if a cer-

tain demand arises or a need is triggered. Agents

can exhibit almost automatic responses that become

habitual. Similarly, if agents have a certain level of

awareness of the impacts of their actions on the en-

vironment, this can lead to a more complex decision

process. Hence, aspects that have an impact on their

decision-making include the agent’s general lifestyle,

preferences, perceived environmental threat posed by

environmental hazards, and group dynamics of their

social network.

https://www.openstreetmap.de/

When looking at related works on agent decision-

making in the context of PEB, one of the focuses is

on the identiﬁcation of decision-relevant factors. For

instance, Tong et al. (2018) investigate under which

circumstances agents decide on an option concern-

ing recycling and waste disposal, Granco et al. (2019)

focus on the conservation and protection of natural

habitats. Mostly, these models utilize psychological

theories to represent the complex decision processes.

One of those is the Theory of Planned Behavior (TPB)

(Ajzen, 1991) that is increasingly used in the con-

text of long-term set goals, i.e., PEB, (cf. Anebagilu

et al. 2021). Personal aspects, like the general atti-

tude towards a behavior or the perceived behavioral

control, affect the formation of an intention to act.

Additionally, the social network plays an important

role in this theory, e.g., by forming behavioral norms

through interactions with the social network. Further-

more, Yuriev et al. (2020) claim that the TPB is well-

suited for the design of behavioral interventions.

To adequately utilize TPB within an agent archi-

tecture, agents must be capable of reﬂecting on their

own situation, desires, goals, and events. For this,

the Belief-Desire-Intention (BDI) model is particu-

larly suitable (Bratman, 1987; Berndt et al., 2018):

individual goals (desires), information (beliefs) and

action-oriented measures (intentions) are organized

into mental states. Intentions are derived from be-

liefs and desires using a deliberation process. TPB

is used to adapt the deliberation process to map the

requirements of the application area by selecting the

necessary internal and external factors for decision-

making.

3.3 Social Layer

Our DT framework acknowledges that the spatial and

individual layers are linked through a social layer, a

complexity often overlooked in DTs. The literature

review showed that important aspects of social orga-

nization and structure are frequently excluded in such

models (cf. Section 2.1).

Literature highlights the pivotal role of social

norms in fostering PEB (Cialdini and Jacobson,

2021), inﬂuencing travel choices (Doran and Larsen,

2016), and speciﬁcally affecting travel mode choices

(Eriksson and Forward, 2011). Travel attitudes of

a region’s inhabitants are signiﬁcant predictors of

travel-related CO

emissions, as exempliﬁed in the

context of the Netherlands (Ettema and Nieuwenhuis,

2017). Other research illustrates the impact of social

learning in virtual environments on the adoption of

pro-environmental lifestyles (Chwialkowska, 2019)

and the uptake of eco-conscious products (Zhang

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

390

et al., 2021). Research therefore suggests that these

concepts are integral to PEB. However, they are often

viewed merely as extensions of the TPB (cf. Eriksson

and Forward 2011; Doran and Larsen 2016; Cialdini

and Jacobson 2021), an approach that risks neglecting

the dynamic interactions within the social sphere.

To better understand the origins and spread of at-

titudes and norms, a clear understanding of the social

structures and network dynamics is essential. This

knowledge is also crucial to effectively model possi-

ble emergent behaviors and to identify challenges in

its adoption. Therefore, we treat the social layer as

a separate level of complexity in our analysis. This

essentially implies consideration of both the local and

the social network. The local network is bound to the

geolocation we described in the spatial layer. The so-

cial network consists of connections and relationships

which are not identiﬁable on the map.

When looking at human networks in inhabited ge-

ographic areas, it is crucial to consider that physi-

cal proximity does not necessarily translate into so-

cial proximity. This is particularly vivid from the re-

search on support networks

. Blokland et al. show

how social networks are susceptible to the effects of

mobility and digitization. The authors challenge tra-

ditional views of social networks, which previously

focused on neighborhood ties, suggesting that these

ties are not solely deﬁned by physical proximity but

are increasingly inﬂuenced by digital interactions and

trans-local mobility (cf. Small 2017).

For us, it is vital to conceptualize the social net-

work in a broader sense, encompassing both digi-

talization and mobility factors. Prior research on

changes in PEB emphasizes the vital role online com-

munication plays on social learning (Chwialkowska,

2019; Zhang et al., 2021). Our DT framework incor-

porates potential pathways for the dissemination of

PEB that have been overlooked in prior studies, due

to their limited scope in deﬁning the social domain.

To consider social networks in our DT framework,

while keeping the individual layer in mind, we fo-

cus on two types of information ﬂow. Agents ac-

quire information from their local and social networks

either through direct observation or via knowledge

and views disseminated by others. This information

is then channeled to the individual layer, and pro-

cessed within the agent’s deliberation. The outcome

results in an intention, tailored to each agent’s spe-

ciﬁc circumstances, that potentially inﬂuences opin-

ion dynamics within their environment. The social

Support network is conceptualized as a system through

which individuals exchange resources, with an emphasis

on the varying strength of ties and the closeness these ties

might represent (Blokland et al., 2021)

layer plays a pivotal role in shaping agents’ percep-

tion of their environment and how their behaviors im-

pact others within the social network.

Incorporating these social aspects into MAS for

the purpose of conducting experiments leads to

the development of Agent-Based Social Simulation

(ABSS) (Davidsson, 2002). ABSS has proven to be

a good ﬁt to enable interaction between agents while

considering social concepts; it offers a controlled en-

vironment for experimenting, e.g., with different sce-

narios or psychological and social concepts (Squaz-

zoni et al., 2014).

4 USE CASE: DIGITAL TWIN OF

A RURAL AREA

This section addresses speciﬁcs of and required data

for our rural area use case. Essentially, the area con-

sists of numerous small communities. Typical for this

region are various demographic changes, i.e., rural

migration and a high proportion of commuters within

and between rural districts (Dauth and Haller, 2018).

Due to poorly developed public transport infrastruc-

ture, people largely depend on private vehicles for

work, leisure activities, and daily supply. To mitigate

the impact of inadequate infrastructure, approaches in

cooperative mobility and logistics are explored.

Figure 2 displays the general processes required

to model the use case. Data Processing and Model

Building shows the sub-models of the layers deﬁned

in Section 3. Theories contains theories applied in the

respective layers. For the spatial layer, we use a net-

work generator to ﬁlter the given data and build the

required networks. Together, these boxes comprise a

generic representation of the DT model. Expanding

this DT using data enables adaptation to the region in

focus and validation of the model components. Data

input and theories are interrelated, e.g., theories de-

termine which data is needed. Lastly, Experimental

Setup and Output are required for simulation studies

that can be conducted using the implemented model.

An experiment consists of testing hypotheses refer-

ring to the speciﬁc scenarios and interventions (Lorig

et al., 2017). The parameter setting depends on the

structure of the model and available data. The output

generally refers to the ratio of adopted behavior by

individuals on the micro and macro levels.

A baseline scenario is the usual shopping behav-

ior where inhabitants order goods online or buy them

from the nearest shop using their preferred transport

option. As an option for PEB, we introduce crowd

delivery supported by commuters. Crowd delivery in

logistics is an approach to unburden delivery services

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas

391

Empirical

Studies

Census

OSM

Data

Individual Layer

Spatial Layer

Social Layer

Decision

Model

Daily

Schedule

Social

Network

Local

Network

Population

Infra-

structure

Theory of Planned

Behavior

Behavioral

Archetypes

Belief-Desire-

Intention

Architecture

Practice Theory

Statistical

Data

Data Input

Data Processing and Model Building

Theories

Social

Mechanisms

Network

Generator (OSM)

Calibration and Validation

Experimental

Setup

Output

Ratio of Adopted Behavior

Parameter

Setting

Knowledge

Hypotheses

Scenarios

Interventions

Figure 2: Processes, Data Use and Theories.

and improve the last mile in terms of parcel delivery

time (cf. Asdecker and Zirkelbach 2020). Inhabitants

are provided with an additional option, i.e., to com-

mission a commuter who travels from, to, or through

the community of the inhabitant. Interventions seek

to encourage residents to utilize this additional option,

for example, by promoting social norms (social pres-

sure) that emerge through interactions or by provid-

ing information about its environmental beneﬁts. The

intended output would be, e.g., an increased ratio of

people adopting the new option that helps to identify

the most effective interventions for real-life applica-

tion.

To build the model components for each DT layer,

a variety of data sources are needed. The spatial layer

requires georeferenced data including buildings and

streets with tags and an overview of public transport.

For this, we use OSM data. While OSM provides a

collection of data for a general area, data for the fo-

cus area has to be extracted by removing all data out-

side the outermost longitude and latitude values of the

area. Of the remaining data, workplaces, and POIs are

ﬁltered using relevant tags. Additionally, we utilize

census data

to generate an artiﬁcial population that

represents the demographic characteristics of respon-

dents, including age, gender, and living situation, at

the time of the survey.

In the following step, inhabitants are represented

by agents with a daily schedule and a decision-

making model. To understand individual aspects, it

is essential to conduct empirical studies, speciﬁcally,

surveys focused on rural areas. These surveys should

https://ergebnisse2011.zensus2022.de/datenbank/

online/

encompass a broad range of topics, including subjec-

tive personal attitudes, preferences, and habits (e.g.,

shopping behavior, travel choices, and daily routines).

Additionally, they need to shed light on the inﬂuences

inspiring PEB, aiming to map out individuals’ social

networks and pinpoint other signiﬁcant external fac-

tors.

An agent’s daily schedule, on the one hand, is in-

ﬂuenced by the person’s workplace. Hence, we in-

clude statistical data regarding commuters in the area

using the Pendleratlas

(Commuter atlas), which tells

how many people commute between and within com-

munities. On the other hand, the schedule is shaped

by leisure activities (e.g., meeting friends), necessary

activities (e.g., grocery shopping), and the respective

travel choices, for which we use empirical studies

and statistical data (e.g., Datenreport Umwelt, En-

ergie und Mobilit

at (Data report environment, energy,

and mobility) by Brockjan et al. (2021)). For the TPB

in the Decision Model we derive data from empiri-

cal studies. Additionally, we utilize archetypes based

on survey results or theoretical constructs, e.g., social

actor types (Dittrich and Kron, 2002).

The local network, deﬁned by data from the spa-

tial layer, includes an individual’s household and can

extend to neighbors. Social networks consist of peo-

ple who might not live in the same neighborhood but

have a strong impact on an individual’s behavior. To

map the social network of our agents, the social net-

work is analyzed based on survey data. Importantly,

our social network analysis centers on the study of

practices (Blokland et al., 2021) rather than prede-

ﬁned ties. This approach reﬂects the ﬂuidity of mod-

ern social interactions and incorporates diverse com-

munication methods.

For validation, we utilize publicly available data

(i.e., commuter statistics, census data, OSM data).

This also facilitates the transferability to other rural

regions. Furthermore, we utilize trafﬁc count data

from the selected region to assess the number of com-

muters and their daily travel routes. The data gathered

from empirical studies is employed to initialize and

calibrate our model and to verify assumptions regard-

ing the model’s behavior.

5 CONCLUSION

This paper aims to understand and foster pro-

environmental behaviors in rural areas. For that, we

ﬁrst conducted a literature review on the theme of

Digital Twins and inhabited areas. We then addressed

https://www.pendleratlas.de/

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

392

current research gaps, and guided by our research

focus, outlined a framework for developing a Digi-

tal Twin. Our framework is distinctive in its multi-

layered approach, integrating spatial, individual, and

social aspects to offer a holistic understanding of the

factors inﬂuencing pro-environmental behavior. Our

framework was created with a real-world use case of a

rural area in mind. We accounted for the unique char-

acteristics of this region and established the basis for

a DT which could facilitate the search of sustainable

solutions, without necessitating signiﬁcant changes to

physical infrastructure. As a next step, we plan to im-

plement the conceptual structure into a computational

model. Our model can support decision-making by

enabling the exploration of various scenarios and in-

terventions that promote pro-environmental behavior.

However, we also acknowledge certain limitations of

this paper. Firstly, our literature review was conﬁned

to a single database, which, while comprehensive,

may have led to the exclusion of relevant publications

from other sources.

Secondly, our approach focused on integrating the

speciﬁc psychological and social theories while omit-

ting others. Future research could beneﬁt from incor-

porating a wider range of theories to capture a more

diverse array of behavioral inﬂuences and dynamics.

Additionally, our conceptualization of the social net-

work has been primarily informed by the research ap-

plicable to urban areas, it is therefore yet to be deter-

mined how well these theoretical considerations will

perform in the rural setting. Third, besides the Digital

Twin’s level of granularity that we addressed in the

spatial layer, other characteristics of DTs to describe

their capability of representing its physical twin can

be discussed in more detail, e.g., ﬁdelity, maturity or

consistency (Su et al., 2023).

Lastly, it is important to consider the poten-

tial variability in the applicability of our framework

across different rural and urban contexts as inhab-

ited areas are diverse in their geographic and socio-

economic characteristics. However, our approach is

promising as it includes the identiﬁcation of key char-

acteristics and establishes basics for an adaptable con-

ceptual framework by building upon openly available

data.

ACKNOWLEDGEMENTS

This work is funded by the Federal Ministry for the

Environment, Nature Conservation, Nuclear Safety

and Consumer Protection (Bundesministerium f

Umwelt, Naturschutz, nukleare Sicherheit und Ver-

braucherschutz) (BMUV, No. 67KI31073A).

REFERENCES

Adreani, L., Bellini, P., Fanfani, M., Nesi, P., and Pantaleo,

G. (2023). Design and develop of a smart city digital

twin with 3d representation and user interface for what-

if analysis. In ICCSA 2023, pages 531–548. Springer.

Agency and Environment, E. (2022). Transport and envi-

ronment report 2022 - Digitalisation in the mobility sys-

tem - Challenges and opportunities. Publications Ofﬁce

of the European Union.

Ahn, C., Ham, Y., Kim, J., and Kim, J. (2020). A digital

twin city model for age-friendly communities: capturing

environmental distress from multimodal sensory data. In

HICCS 2020.

Ajzen, I. (1991). The theory of planned behavior. Or-

ganizational behavior and human decision processes,

50(2):179–211.

An, L. (2012). Modeling human decisions in coupled hu-

man and natural systems: Review of agent-based models.

Ecological Modelling, 229:25–36. Modeling Human De-

cisions.

Anebagilu, P. K., Dietrich, J., Prado-Stuardo, L., Morales,

B., Winter, E., and Arumi, J. L. (2021). Application of

the theory of planned behavior with agent-based model-

ing for sustainable management of vegetative ﬁlter strips.

Journal of Environmental Management, 284:112014.

Asdecker, B. and Zirkelbach, F. (2020). What drives the

drivers? a qualitative perspective on what motivates the

crowd delivery workforce. In HICSS 2020, pages 1–10.

Barat, S., Paranjape, A., Parchure, R., Darak, S., and Kulka-

rni, V. (2022). Agent based simulatable city digital twin

to explore dynamics of covid-19 pandemic. In WSC

2022, pages 557–568. IEEE.

Bellini, P., Bilotta, S., Palesi, A. L. I., Nesi, P., and Pantaleo,

G. (2022). Vehicular trafﬁc ﬂow reconstruction analysis

to mitigate scenarios with large city changes. IEEE Ac-

cess, 10:131061–131075.

Berndt, J. O., Rodermund, S. C., and Timm, I. J. (2018).

Social contagion of fertility: An agent-based simulation

study. In WSC 2018, pages 953–964. IEEE.

Blokland, T., Kr

uger, D., Vief, R., and Schultze, H. (2021).

Where we turn to: Rethinking networks, urban space,

and research methods. In Spatial Transformations, pages

258–268.

Bonabeau, E. (2002). Agent-based modeling: Methods and

techniques for simulating human systems. Proc. of the

national academy of sciences, 99(suppl

3):7280–7287.

Bratman, M. (1987). Intention, Plans, and Practical Rea-

son. Cambridge, MA: Harvard University Press, Cam-

bridge.

Brockjan, K., Maier, L., Kott, K., and Seqald, N. (2021).

Umwelt, Energie und Mobilit

at. Auszug aus dem Daten-

report 2021. Technical report, Statistisches Bundesamt.

Bujari, A., Calvio, A., Foschini, L., Sabbioni, A., and Cor-

radi, A. (2021). A digital twin decision support sys-

tem for the urban facility management process. Sensors,

21(24):8460.

Chapman, L. (2007). Transport and climate change: a re-

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas

393

view. Journal of transport geography, 15(5):354–367.

Chwialkowska, A. (2019). How sustainability inﬂuencers

drive green lifestyle adoption on social media: The pro-

cess of green lifestyle adoption explained through the

lenses of the minority inﬂuence model and social learn-

ing theory. Management of Sustainable Development,

11(1):33–42.

Cialdini, R. B. and Jacobson, R. P. (2021). Inﬂuences of so-

cial norms on climate change-related behaviors. Current

Opinion in Behavioral Sciences, 42:1–8.

Clemen, T., Ahmady-Moghaddam, N., Lenfers, U. A.,

Ocker, F., Osterholz, D., Str

obele, J., and Glake, D.

(2021). Multi-agent systems and digital twins for smarter

cities. In ACM SIGSIM-PADS 2021, pages 45–55.

Dauth, W. and Haller, P. (2018). Beruﬂiches Pendeln zwis-

chen Wohn-und Arbeitsort: Klarer Trend zu l

angeren

Pendeldistanzen. Technical report, IAB-Kurzbericht.

Davidsson, P. (2002). Agent based social simulation: A

computer science view. Journal of artiﬁcial societies and

social simulation, 5(1).

Dawes, R. M. (1980). Social dilemmas. Annual review of

psychology, 31(1):169–193.

Dittrich, P. and Kron, T. (2002). Complex reﬂexive agents

as models of social actors. In Proc. of the SICE Workshop

on Artiﬁcial Society/Organization/Economy, volume 25,

pages 79–88.

Doran, R. and Larsen, S. (2016). The relative importance

of social and personal norms in explaining intentions to

choose eco-friendly travel options. International Journal

of Tourism Research, 18(2):159–166.

Eriksson, L. and Forward, S. E. (2011). Is the intention to

travel in a pro-environmental manner and the intention

to use the car determined by different factors? Trans-

portation Research Part D: Transport and Environment,

16(5):372–376.

Ettema, D. and Nieuwenhuis, R. (2017). Residential self-

selection and travel behaviour: What are the effects of

attitudes, reasons for location choice and the built envi-

ronment? Journal of Transport Geography, 59:146–155.

Fan, Z., Yang, X., Yuan, W., Jiang, R., Chen, Q., Song, X.,

and Shibasaki, R. (2022). Online trajectory prediction for

metropolitan scale mobility digital twin. In SIGSPATIAL

2022, pages 1–12.

Ferreira, N., Poco, J., Vo, H. T., Freire, J., and Silva, C. T.

(2013). Visual exploration of big spatio-temporal ur-

ban data: A study of new york city taxi trips. IEEE

Transactions on Visualization and Computer Graphics,

19(12):2149–2158.

Granco, G., Heier Stamm, J. L., Bergtold, J. S., Daniels,

M. D., Sanderson, M. R., Sheshukov, A. Y., Mather,

M. E., Caldas, M. M., Ramsey, S. M., Lehrter II, R. J.,

Haukos, D. A., Gao, J., Chatterjee, S., Nifong, J. C., and

Aistrup, J. A. (2019). Evaluating environmental change

and behavioral decision-making for sustainability policy

using an agent-based model: A case study for the smoky

hill river watershed, kansas. Science of The Total Envi-

ronment, 695:133769.

Grieves, M. (2014). Digital Twin: Manufacturing Excel-

lence through Virtual Factory Replication. Technical re-

port, Michael W. Grieves, LLC.

Helbing, D. (2012). Social self-organization: Agent-based

simulations and experiments to study emergent social be-

havior. Springer.

Le Fur, J., Sall, M., and Dembele, J.-M. (2023). A Digital

Twin Simulator Approach as a Support to Develop an

Integrated Observatory of the Epidemic Risk in a Rural

Community in Senegal. In SIMULTECH 2023, pages

134–142. SCITEPRESS.

Leiserowitz, A., Carman, J., Buttermore, N., Neyens, L.,

Rosenthal, S., Marlon, J., Schneider, J., and Mulcahy, K.

(2022). International public opinion on climate change

2022.

Lorig, F., Lebherz, D. S., Berndt, J. O., and Timm, I. J.

(2017). Hypothesis-driven experiment design in com-

puter simulation studies. In WSC 2017, pages 1360–

1371. IEEE.

Major, P., Li, G., Hildre, H. P., and Zhang, H. (2021). The

use of a data-driven digital twin of a smart city: A case

study of

alesund, norway. IEEE Instrumentation & Mea-

surement Magazine, 24(7):39–49.

Major, P., Torres, R., Amundsen, A., Stadsnes, P., et al.

(2022). On the use of graphical digital twins for urban

planning of mobility projects: a case study from a new

district in

alesund, norway. In ECMS 2022.

Masson-Delmotte, V., Zhai, P., P

ortner, H.-O., Roberts, D.,

Skea, J., Shukla, P. R., et al. (2022). Global Warming of

1.5 C: IPCC special report on impacts of global warming

of 1.5 C above pre-industrial levels in context of strength-

ening response to climate change, sustainable develop-

ment, and efforts to eradicate poverty. Cambridge Uni-

versity Press.

Mavrokapnidis, D., Mohammadi, N., and Taylor, J. (2021).

Community dynamics in smart city digital twins: A com-

puter vision-based approach for monitoring and forecast-

ing collective urban hazard exposure. In HICSS 2021.

Meta, I., Serra-Burriel, F., Carrasco-Jim

enez, J. C., Cucchi-

etti, F. M., Div

ı-Cuesta, C., Garc

ıa Calatrava, C., Garc

ıa,

D., Graells-Garrido, E., Navarro, G., L

azaro, Q., et al.

(2021). The camp nou stadium as a testbed for city phys-

iology: a modular framework for urban digital twins.

Complexity, 2021:1–15.

Mohammadi, N. and Taylor, J. (2019). Devising a Game

Theoretic Approach to Enable Smart City Digital Twin

Analytics. In HICCS 2019.

Mohammadi, N. and Taylor, J. (2020). Knowledge discov-

ery in smart city digital twins. In HICSS 2020.

National Academies of Sciences, E., Medicine, et al.

(2020). Climate change: evidence and causes up-

date 2020. The National Academies Press, Washington,

17226:25733.

Pan, X., Mohammadi, N., and Taylor, J. E. (2022). Smart

city digital twins for public safety: A deep learning

and simulation based method for dynamic sensing and

decision-making. In WSC 2022, pages 808–818.

Shukla, P. R., Skea, J., Calvo Buendia, E., Masson-

Delmotte, V., P

ortner, H. O., Roberts, D., Zhai, P., Slade,

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

394

R., Connors, S., Van Diemen, R., et al. (2019). Ipcc,

2019: Climate change and land: an ipcc special report

on climate change, desertiﬁcation, land degradation, sus-

tainable land management, food security, and greenhouse

gas ﬂuxes in terrestrial ecosystems.

Small, M. L. (2017). Someone to Talk to. Oxford University

Press, New York.

Sottet, J.-S., Brimont, P., Feltus, C., G

ateau, B., and

Merche, J. F. (2022). Towards a lightweight model-

driven smart-city digital twin. In MODELSWARD, pages

320–327.

Squazzoni, F., Jager, W., and Edmonds, B. (2014). Social

simulation in the social sciences: A brief overview. So-

cial Science Computer Review, 32(3):279–294.

Su, S., Nassehi, A., Hicks, B., and Ross, J. (2023). Charac-

terisation and evaluation of identicality for digital twins

for the manufacturing domain. Journal of Manufacturing

Systems, 71:224–237.

uddeutsche Zeitung (2022). Mehrheit der Men-

schen w

urde h

auﬁger

OPNV nutzen. https://www.

sueddeutsche.de/wirtschaft/verkehr-mehrheit-der-

menschen-wuerde-haeuﬁger-oepnv-nutzen-dpa.urn-

newsml-dpa-com-20090101-220111-99-665040. 2022-

01-11.

Tong, X., Nikolic, I., Dijkhuizen, B., van den Hoven, M.,

Minderhoud, M., W

ackerlin, N., Wang, T., and Tao,

D. (2018). Behaviour change in post-consumer recy-

cling: applying agent-based modelling in social exper-

iment. Journal of Cleaner Production, 187:1006–1013.

Van Den Berghe, S. (2021). A processing architecture for

real-time predictive smart city digital twins. In IEEE Big

Data 2021, pages 2867–2874. IEEE.

VanDerHorn, E. and Mahadevan, S. (2021). Digital

twin: Generalization, characterization and implementa-

tion. Decision Support Systems, 145:113524.

Wang, H., Chen, X., Jia, F., and Cheng, X. (2023). Digital

twin-supported smart city: Status, challenges and future

research directions. Expert Systems with Applications,

page 119531.

Wohlin, C. (2014). Guidelines for snowballing in system-

atic literature studies and a replication in software engi-

neering. In EASE 2014, pages 1–10.

Yun, T.-S., Park, M., and Moon, I.-C. (2023). Modeling and

calibrating digital twin of automatic garbage collection

system in sejong city. In ANNSIM 2023, pages 572–583.

IEEE.

Yuriev, A., Dahmen, M., Paill

e, P., Boiral, O., and Guillau-

mie, L. (2020). Pro-environmental behaviors through the

lens of the theory of planned behavior: A scoping review.

Resources, Conservation and Recycling, 155:104660.

Zhang, W., Chintagunta, P. K., and Kalwani, M. U. (2021).

Social media, inﬂuencers, and adoption of an eco-

friendly product: Field experiment evidence from rural

china. Journal of Marketing, 85(3):10–27.

Driving Towards a Sustainable Future: A Multi-Layered Agent-Based Digital Twin Approach for Rural Areas

395