Cities 110 (2021) 103064

Contents lists available at ScienceDirect

Cities
journal homepage: www.elsevier.com/locate/cities

A digital twin smart city for citizen feedback

Gary White *, Anna Zink, Lara Codecá, Siobhán Clarke
Trinity College Dublin, College Green, Dublin 2, Ireland

A R T I C L E I N F O A B S T R A C T

Keywords: A digital twin is a digital representation of a physical process, person, place, system or device. Digital twins were
Digital twin originally designed to improve manufacturing processes using simulations that have highly accurate models of
Smart cities individual components. However, with increasingly large and accurate building information models (BIM)
Internet of things (IoT)
combined with big data generated from IoT sensors in a smart city, it is now possible to create digital twin smart
Urban planning
Urban policy
cities. An accurate 3D model of a city can be published online and walked around by the public to view proposed
changes in urban planning and policy. This allows for easier dissemination and transparency to the public before
putting these decisions into practice. This open and public model allows for an additional virtual feedback loop
where citizens can interact and report feedback on planned changes in the city. Citizens can also interact with
components to tag and report problems in their area. The digital twin also allows for additional experimentation
where 3D data is necessary, such as flood evacuation planning. In this paper, we demonstrate a public and open
digital twin of the Docklands area in Dublin, Ireland and show how this model can be used for urban planning of
skylines and green space allowing users to interact and report feedback on planned changes.

1. Introduction (Zhao et al., 2019). In the digital twin layer reinforcement learning al­
gorithms can also be used to learn the best action policies to improve
Cities have become increasing smarter in the last two decades (Al­ performance in a number of urban intelligence tasks, such as managing
bino et al., 2013), using pervasive information and communications traffic and power systems (Hsu et al., 2014; Mannion et al., 2016).
technology (ICT) to monitor activities in the city (Neirotti et al., 2014). The increased data available from smart cities, artificial intelligence,
Data can then be generated from a wide variety of activities in the city, data analytics and machine learning allows for the creation of a digital
such as traffic and transportation (Menouar et al., 2017), power gener­ twin that can update and change as the physical equivalents change
ation (Oldenbroek et al., 2017), utilities provisioning (Sánchez et al., (Kaur et al., 2020). A digital twin is a pairing of the virtual and physical
2013), water supply (Parra et al., 2015) and waste management (Med­ worlds that allows for analysis of data and monitoring of systems to head
vedev et al., 2015). Smart cities can then use this data to improve the off problems before they occur, prevent downtime and can even be used
mobility, environment, living standards and governance of the city to plan for the future using simulations (Boschert & Rosen, 2016).
(Abella et al., 2017; Angelidou, 2015). A strong link has been shown Digital twins have primarily been used in the manufacturing sector, but
between investing in smart city policies and urban GDP growth (Car­ other areas of study and business are beginning to find new potential
agliu & Del Bo, 2019). uses. An ideal digital twin would be identical to its physical counter-part
The data generated by smart cities makes them exciting testbeds for and have a complete, real-time dataset of all information on the object/
data mining and machine learning (Mohammadi & Al-Fuqaha, 2018; system. As the object/system increases in complexity a digital twin may
White & Clarke, 2020). The services provided to citizens in a smart city be identical in only relevant areas and have only the real-time data
can be personalised using machine learning, internet of things and big necessary to support any desired simulations. How accurate and useful a
data (Chin et al., 2017; White, Palade, et al., 2019). These deep learning digital twin is, depends on the level of detail put into it and how
algorithms can be used to categorise and perform analytics on a number comprehensive the available data is.
of different data streams including videos (Wang & Sng, n.d.). More Digital twins allow for the simulation of many options before taking
recent neural network approaches, such as generative adversarial physical action in the real world to identify the strengths and weak­
network (GAN) can be used to optimize crowd routing in a smart city nesses of each plan. This is especially important in safety critical

* Corresponding author.
E-mail addresses: whiteg5@scss.tcd.ie (G. White), zinka@tcd.ie (A. Zink), lara.codeca@tcd.ie (L. Codecá), siobhan.clarke@scss.tcd.ie (S. Clarke).

https://doi.org/10.1016/j.cities.2020.103064
Received 24 January 2020; Received in revised form 17 June 2020; Accepted 27 November 2020
Available online 8 January 2021
0264-2751/© 2020 Elsevier Ltd. All rights reserved.
G. White et al. Cities 110 (2021) 103064

situations, where only one option can be chosen and there may be a used in a number of complex systems beyond product design and the
number of competing plans to choose from. This is exemplified by the manufacturing process. Digital twins can be used to create digital twin
rescue operation in Thailand to save a lost soccer team that occurred in humans that can be used in healthcare (Bruynseels et al., 2018). With
July 2018 (Puri, n.d.) (Dixon, n.d.). A 3D map of the terrain, a complex the rise of quantified-self, users can now collect more data about their
cave system, was created using GIS data, water and oxygen information physical activity, sleep quality, diet, heart rate, weight, productivity,
inside the caves. Weather forecasts were also used in order to create an working environment and social interaction (White, Liang, et al., 2019).
accurate digital twin that could simulate rescue operations and ensure This data can then be used to create an accurate digital twin to predict
the safety of the rescuers and the lost team. The use of a digital twin upcoming health issues as well as test solutions to prevent or reduce the
ensured that when rescuers acted, it was a best-case scenario after damage of any complications (Bhavnani & Sitapati, 2019).
testing multiple options. Digital twin cities can be created using the data collected from smart
Digital twins can have applications in a number of different domains. city services (Mohammadi & Taylor, 2017). The virtual representation
With the data generated by smart cities, digital twins can be used to allows for modelling and visualisation of the spatiotemporal information
model urban planning and policy decisions. An example of a work-in- in a city. Much of the recent success in smart cities around the world in
progress digital twin of a city is Virtual Singapore,1 which is a three- integrating reliable ICT systems into the city can be utilised to create a
dimensional (3D) city model and data platform (Alam & El Saddik, digital twin of a city (Mohammadi & Taylor, 2019; White et al., 2017).
2017). In this paper, we expand on the digital twin ideas introduced in An initial attempt to create a digital twin smart city has been conducted
Virtual Singapore and make our model publicly available on the in Singapore, also known as Virtual Singapore (Soon & Khoo, 2017).
internet.2 This allows users to easily interact with the model, leave However, there are a number of limitations with this initial approach as
feedback about urban planning decisions and tag problems in their local the model has not been made publicly available, so citizens cannot
area. This generates an additional layer of data that can be used to make interact with the model or report feedback and it does not include urban
changes in a smart city. We also show how the 3D model can be used to mobility data. Digital twin programmes are in the early stages with a
create realistic flooding and crowd simulations. roadmap being outlined at the Centre for Digital Built Britain at Cam­
In this paper, we present an open and publicly available digital twin bridge University (Enzer et al., n.d.). The roadmap shows the key
smart cities model of the Docklands area in Dublin, Ireland. Section 2 building blocks and actors that together would enable successful digital
presents the related work that has been conducted using digital twins in twins across the built environment. In this paper we tackle point 3.10 to
different domains. Section 3 presents the design of the digital twin smart undertake strategic pilots to prove the information architecture with
city model and how it is created through the combination of data from a selected stakeholders and point 4.7 to share learning from digital twin
number of different layers across the city. Section 4 presents the hubs, pilots and demonstrations.
experimental setup that was used to conduct the simulations and the A number of private companies, such as CityZenith,3 Agency9 who
feedback that was received by making the model publicly available. were acquired by Bentley4and SmarterBetterCities5 have started to
Section 5 presents the results of those simulations and Section 6 presents develop in the Digital Twin Smart City space. However, all of these
some of the limitations of the current digital twin model and how it companies are private and do not make their models publicly available
could be improved in future iterations. Section 7 concludes the paper and charge expensive licensing fees. Our approach is publicly released
and presents some future work. on the internet allowing users to interact with the model and for it to be
used in future digital twin projects for free.
2. Related work In this paper, we create a digital twin of the Docklands area in
Dublin, Ireland using a publicly released 3D model.6 We show how the
Digital twins are a digital replica of a living or non-living physical model can be used for a number of urban planning and policy decisions
entity (El Saddik, 2018). Enabling tools for digital twin data, services, by adding a proposed building to the skyline as well as additional green
modelling and connection to the physical world have led to the spaces and parks. As the model is available online1 users can easily tag
increased popularity of digital twins (Qi et al., n.d.). Digital twins problem areas in the city and fill forms to make changes in their local
integrate internet of things, machine learning, artificial intelligence and area. This generates additional data that can feed back into the digital
data analytics to create living digital simulation models that update and twin to specify problem areas in the city that need to be developed. We
change with their physical counterparts (Luo et al., 2019). A digital twin also show how our digital twin can be used to create a number of urban
is continuously learning and updating itself from multiple data sources mobility simulations using pedestrian mobility, as well as simulating the
to represent the physical object in near real-time. The system can learn effect that the river flooding would have on the city.
from itself, from other similar digital twins or from human experts with
relevant domain knowledge. A digital twin can also learn from historical 3. Digital twin smart city design
data in past usage and factor this into its digital model.
Digital twins were first defined by NASA as a paradigm for future A digital twin smart city builds on a number of layers of information
NASA and U.S Air Force vehicles (Glaessgen & Stargel, 1818). A digital in the city. We define six layers in our digital twin smart city model, as
twin would allow for ultra-high fidelity simulation using data from the shown in Fig. 1. The first five layers build on top of each other adding
vehicle’s on-board system, maintenance history and all available his­ more information about the terrain, buildings, infrastructure, mobility
torical and fleet data to identify any possible problems in safety or and IoT devices in the city. The Digital Layer/Smart City is used to
reliability. They have since been applied to a number of manufacturing collect data from the city, which it can then pass to the Virtual Layer/
projects as they can bridge the gap between the virtual and physical Digital Twin. The Digital Twin uses the data generated in the smart city
space at different stages in the product’s lifetime (Tao et al., 2018). A to perform additional simulations about mobility optimisation, building
digital twin allows for the product to be tested at all stages of the design placement or the design of renewable energy such as offshore wind
process to ensure the design is feasible, safe, efficient and reliable (Rosen turbines. This information is then fed back through the layers of the
et al., 2015).
A digital twin makes control and experimentation of a complex
system feasible (Grieves & Vickers, 2017). This has led to them being 3
https://cityzenith.com/.
4
https://www.bentley.com/en/products/product-line/reality-modeling-soft
ware/opencities-planner.
1 5
https://www.nrf.gov.sg/programmes/virtual-singapore. https://www.smarterbettercities.ch/.
2 6
https://www.scss.tcd.ie/~whiteg5/webgl/. https://data.smartdublin.ie/dataset/3d-data-hack-dublin-resources.

2
G. White et al. Cities 110 (2021) 103064

model where it is implemented in the physical world. In this section we

describe each layer of the model as follows:

3.1. Terrain

The zeroth layer of the digital twin smart city design is the terrain on
which the city is built. This is basic information about the city, such as
5-Virtual Layer what part of the city is offshore, are there rivers or canals running
/Digital Twin through the city, are there steep gradients or hills in the city, what part
of the city is made of sand, what area of the city has fertile soil that can
Information be used to grow crops, what areas of the city have soil with poor draining
that can cause problems, such as a landslide or during heavy rain or
Data flooding. A soil map, as show in Fig. 2 can be used to incorporate this
information in the model.

3.2. Buildings
4-Digital Layer
/Smart City Layer 1 of the digital twin smart city model then adds the current
buildings in the city to the model. These buildings have highly accurate
building information modelling (BIM) models that can be used as a
digital twin of the building. The 3D building data can also be generated
using stereoscopic aerial photography. The building data used in our
digital twin is of the Dublin Docklands district in Ireland. This data is
publicly available as an FBX file2.
The data is limited to the area between the Samuel Beckett Bridge
and the Eastlink Bridge. It contains the region two blocks north of the
3-Mobility river and partially contains the region south of the river up to the end of
the Grand Canal.

3.3. Infrastructure

Layer 2 of the digital twin smart city model then adds the infra­
structure that surrounds the current buildings in the city. This is the
basic physical and organisational structures and facilities (e.g. roads,
power supplies, telecommunication) needed for the operation of a so­
ciety or enterprise. This infrastructure data can come from Open­
2-Infrastructure StreetMap, which contains information about power, public transport,
motorways, highways, amenities and telecoms. Data can also come from
the 3D mapping process to add gradient information as this may not be
available in open street maps.

3.4. Mobility

Layer 3 of the digital twin smart city model adds mobility to the
infrastructure and building layers. Mobility is the movement of people
during their daily routing and the movement of the goods that help them
1-Buildings
in different aspects of their lives. Software applications such as SUMO
can be used to simulate urban mobility (Lopez et al., 2018). SUMO
simulator supports a number of different transportation modes: walking,
bicycles, powered two-wheelers and generic parametrized vehicles.
Additional available models are railways and waterways. This applica­
tion can be connected to the 3D model in Unity using the Traffic Control
Interface (TraCI). Unity can also be used to implement and enhance the
traffic modes, adding additional behaviours that the simulator does not
model. In our case, we are interested in adding multiple pedestrian
0-Terrain types, such as adult and elderly.

3.5. Digital layer/smart city

Fig. 1. Layers required to develop a digital twin smart city. The digital layer/smart city layer has become hugely popular with a
number of projects focused on integrating IoT sensors in the city to
collect data (Dameri et al., n.d.; Caragliu et al., 2011; Cocchia, 2014).
This data can then be used to monitor and manage traffic and trans­
portation systems (Menouar et al., 2017), power plants (Oldenbroek
et al., 2017), utilities (Sánchez et al., 2013), water supply networks
(Parra et al., 2015), waste management (Medvedev et al., 2015), crime

3
G. White et al. Cities 110 (2021) 103064

Fig. 2. Soil map.

detection (Chiodi, 2016), information systems (Abdel-Basset & components from each of the layers. For example, we do not need the
Mohamed, 2018), schools (Williamson, 2015), libraries (Johnson, n.d.), city terrain information, city buildings, urban traffic mobility model or
hospitals (Pramanik et al., 2017), and other community services 5G telecommunications model to run this simulation. We only use the
(Hashimoto et al., 2015; Jalali et al., 2015). relevant wind, offshore boat and feedback data from the citizens to
As shown in Fig. 1 layer 4, the digital layer/smart city is responsible decide on the size and placement of the offshore wind turbines.
for gathering all the data needed for simulations in the virtual layer/ The digital twin can also be used to aid with the construction of
digital twin, from all the previous layers. The results of these simulations buildings in the city. In the digital twin layer in Fig. 1 we can see that
are then fed back through the layers of the city as information. The data two new large buildings are being proposed: a circular skyscraper and
can come from citizens, devices and assets that may be mobile and the building with a spire on top. Using the sensing data collected in the
located throughout the city. Citizens can use their mobile phones and digital layer, simulations can be created in the virtual layer to see how
smart watches to report data to the city authority. In Fig. 1 layer 4, there these buildings will affect the sunlight in the city e.g., would they block
are connected vehicles that can report traffic data to the traffic authority the sun from existing parks. Wind and seismic data collected in the
to help schedule lights to optimize the flow of traffic. There are a number digital layer can also be used in the design of the new buildings. Once a
of smart trees that have moisture sensors at the roots to ensure that they digital twin of the new buildings has been created they can be tested
are watered at the correct interval. There are a number of connected against the known challenges of the city, such as high winds or being
CCTV cameras located throughout the city as a deterrent for crime and near an earthquake fault line. Once the buildings have attained the
also to be used as evidence for criminal behaviour. There are a number appropriate building safety certificate they can be added to an online
of 5G cell towers located throughout the city that provide fast and low digital twin.
latency access to the internet. A connected fire device has just sent an The online digital twin allows citizens to easily walk around and give
alert to the fire brigade from one of the six tower buildings in the city feedback on new urban policy and planning decisions. Citizens can enter
after detecting smoke in one of the apartments. Wind sensors have also forms to give feedback on newly proposed buildings or green spaces,
been deployed near the monument offshore as well as on the boats to such as parks in the city. This allows the digital twin to generate addi­
collect data for the possible design of an offshore wind farm. tional smart city data that can be used to create information through
experimentation, which is fed back through the layers of the smart city.
For example, in the green space simulation in Section 5, citizens could
3.6. Virtual layer/digital twin
propose new items that they wanted to be included in a new park, which
led to the inclusion of a children’s play pen option, which was not one of
The virtual layer/digital twin builds on the data that is produced
the original proposed options. Fig. 3 shows the online digital twin
from the digital layer/smart city. There is a connection between the
interaction diagram. We can see how the model being deployed online
virtual layer and digital layer as shown by the arrows in Fig. 1. Data is
over the internet allows for easy citizen feedback. This feedback can
sent from the digital layer about the mobility, infrastructure, buildings
then be send to the relevant group, such as the researchers who devel­
and terrain in the city. This data is used to conduct simulations in the
oped the urban mobility model or the city council, who provide the
virtual layer, which can then be passed back as information through the
urban IoT data. The city council can then use this data to make informed
layers of the city. For example, in Fig. 1 wind data is being collected
decisions about urban planning and policy. The main stakeholders for
offshore at the digital layer. This data is then used to conduct simula­
each of the components of the online digital twin are also shown with
tions on the viability of using offshore wind turbines to meet the
the blue box indicating the main technologies used.
renewable energy targets for the city. A highly realistic digital twin of
the turbines can be developed using the wind data collected in the
4. Experimental setup
digital layer to influence the size and placement of the turbines. This
digital twin can then also be used to evaluate the visual impact of the
4.1. Simulation software
turbine placement as citizens may not want the turbines placed close to
the offshore monument. Additional tracking information from the
Unity3D Software (Unity), version 2019.2.10f1 Personal, was used to
offshore boats can also be used to help place the turbines in an area that
load the digital twin, which is a 3D FBX model that contains the first
will not affect offshore traffic. In this case we are only using the relevant

4
G. White et al. Cities 110 (2021) 103064

disapprove of the new building.

Citizens
4.3. Green space simulation
Online Digital
Twin Green space, such as parks and recreational spaces are hugely
important in smart cities for promoting healthy living and wellbeing
WebGL
(Anguluri & Narayanan, 2017; Lee et al., 2015). The digital twin model
Deploy Feedback Internet allows for development of these green spaces in suitable areas of the city.
Data from the smart city, such as air pollution, noise pollution, pedes­
Digital Twin trian traffic flow and amount of direct sunlight can be used to influence
urban planning decisions of where to place these green spaces in the city.
Unity
Model Feedback Open Source If the park or green space gets urban authority approval then the
Data smart city can be used to track the number of people that visit the new
green space. Different facilities in the park, such as extra benches or new
Developed Additonal Citizen
3D and Urban
Model Data
IoT Data flowers can be modelled using the digital twin, while also presenting a
Mobiliy Model
IoT Devices selection of options for users to choose or propose new suggestions. The
Stereoscopy and
SUMO success of this additional equipment can also be tracked through the use
City Council of sensors in the smart city. In our simulations we add additional tree
Researchers and
locations throughout the city and create a new park. Citizens can leave
Model Developers
feedback about what additional equipment they would like to see in the
Fig. 3. Online digital twin interaction diagram. new park by selecting from the following options: more benches, more
summer flowers, park gym equipment or other. By selecting the ‘other’
three layers of terrain, buildings and infrastructure. Pedestrian mobility option the citizen can leave a suggestion that is not one of the considered
models were implemented in Unity to allow for crowd simulations, with options, which may provide additional information of the local citizens
different agents types, including adult and elderly using data from an needs for urban planners.
experimental study (Oxley et al., 2005). Smart city data is taken from the
Dublinked7 site to make the simulations as accurate as possible. The 4.4. User tagging simulation
twining with real public data from an open portal allows the model to
evolve over time and allows for multidisciplinary modelling of the city The initial form-based skyline and green space simulations can be
(Castelli et al., 2019). A city is always evolving over time with new extended to allow citizens to walk around the model more freely and tag
buildings and data available from the city about crowds in the city or objects within it. With an accurate digital twin available, citizens would
flooding information. This new data creates a feedback loop allowing be able to tag real life problems in the model and then have the infor­
the digital twin to test the models and predictions on unseen data. mation sent to the relevant government department. This message
would have the exact location of the issue as seen in the model.
The digital twin can also be used in disaster scenarios to easily show
4.2. Skyline simulation citizens in the city the areas will be most affected by flooding and the
roads that will become inaccessible. Citizens can also tag their current
A version of the model has also been made available online.2 A 3D location in the digital twin and whether or not they are in need of
model allows for the easy removal and addition of newly proposed assistance. This user tagging would enhance the data generated by the
buildings. This can be useful for conducting simulations to evaluate the smart city and would allow for more detailed simulations when using
response to new buildings being introduced in the city and how they the digital twin in future as the locations of where most citizens needed
would affect the skyline. Any proposed building plans can easily be assistance would be captured. The actual level that the water rose based
added to the digital twin using the BIM model. This model would then on the amount of rain would also then be fed back into the digital twin to
allow citizens and public officials to walk around the digital twin and see ensure that the model has an accurate model for predicting the water
the effect that the new building would have on the skyline from a level based on rainfall.
number of different locations. The sunlight information from the smart
city combined with the BIM file of the new building could be combined 4.5. Flooding simulation
to simulate the effect that a large building would have on the sunlight
access of nearby parks or public spaces. Accurate flooding simulations require the detailed 3D terrain and
Building works are long term projects and can take years to com­ building information provided in a digital twin. Unity allows the loading
plete. This can lead to a large difference between the current view of the of a 3D digital twin, which contains accurate street elevation informa­
city and what the city will look like when all the current building pro­ tion. This allows the water level to be dynamically altered, showing the
jects are finished. In our simulation, we show a current view of the areas that would first be affected by a river flooding its banks or heavy
model with all the current buildings finished compared to the current rain. These simulations can be useful for urban authorities in flood
view in the street to highlight this point. We also add some additional planing by deciding where to deploy sandbags and what areas of the city
BIM models as assets to our digital twin to show how the skyline would to evacuate first.
be affected by new buildings. In the digital twin model that is deployed Data from the smart city about rainfall amount and river levels can
online, citizens can be presented with a form that they can use to give be fed to the digital twin to create a timeline of when flooding could
feedback on the newly proposed building. The form asks for the citizens occur. This information can then be passed from the digital twin back to
name and email for verification, then the user can vote on whether or the smart city to alert citizens of the timeline for a possible flood. The
not they approve of the new building. Once the user has voted on historical data collected from the smart city can then be used to create
whether or not they approve of the new building they can leave addi­ longer term flood prevention mechanisms if flooding is identified as a
tional feedback in a text box explaining the reasons why they approve or problem for the city. This could be for large scale projects, such as
introducing water storage areas or diverting rivers. Previous flooding
simulation approaches have integrated propitiatory flooding models,
7
https://data.smartdublin.ie/. such as Mike Flood and have focused on the economic cost caused by

5
G. White et al. Cities 110 (2021) 103064

floods (Pyatkova et al., 2019). Our approach is focused on urban

mobility and identifying the areas that are susceptible to flooding in the
city.
In this flooding simulation we raise the water level in the River Liffey
in Dublin to identify the surrounding areas that would be most affected
by the river flooding its banks. This simulation is able to identify roads
and walkways that would become inaccessible to vehicles and pedes­
trians in a flood, which would alter the movement patterns in the city.
This is especially important for one way roads where vehicles may
become stuck and not be able to turn around. Our initial model is
simplistic, only taking into account rainfall and river levels, but it can be
extended to include additional information about currents, materials in
the river beds, sewage and drainage systems. Our model is validation
using data from the office of public works, which has historical data on
past floods in Ireland and gives low, medium and high probability rat­
ings to a current area being flooded.

4.6. Crowd simulation

The crowd simulation can be carried out anywhere in the model, but
was mostly focused around an intersection just north of the Samuel
Beckett bridge. Spawn locations were placed at entry points to the maps,
as a general collection at the four corners of the main intersection and at
the exits of the 3Arena. These sets of spawn locations are the children of
empty objects that describe the set and any of the sets can be used for a
flock to set the spawn locations and destinations. The average waiting
time and the average distance travelled was then compared between the
agent types. The two agent types elderly and adult, were spawned at
different ratios: 0–100, 25–75, and 50–50 splits. The Elderly agents have
Fig. 4. Approved building works.
a smaller maximum speed range, step height and wider radius so that
they would not make as sharp turns, which is based on an experimental
evaluation (Oxley et al., 2005). Their average waiting times and distance view of the Docklands available on Google Maps and the current simu­
travelled were then compared. lated model. We can see that there is a large difference between the
A flocking algorithm was developed to control pedestrian movement. current view in Fig. 4a and the model view in Fig. 4b, with many
This is done by assigning behaviours to the pedestrians, which will be buildings still under construction. The digital twin gives citizens easy
referred to as agents. These behaviours determine how the agent will
move when interacting with its environment and other agents. The core
of this algorithm contains three scripts, FlockBehaviour, FlockAgent,
and Flock. The first, FlockBehaviour is a scriptable object, which all
behaviour scripts inherit from. Each behaviour creates a movement
vector that are summed to find the agents next move. The second and
third are both Monobehaviours that govern the properties of the flock.
FlockAgent controls individual properties and stores data that is unique
to the individual agent, while Flock controls anything that is universal to
the flock. The three basic behaviours needed are cohesion, alignment
and avoidance. These ensure that the flock of agents move together, in
the same direction, and that the agents do not overlap with each other.
Other behaviours have been added to make a more realistic crowd
simulation or to incite reactions to any introduced stimuli. The flocking
algorithm is made publicly available.8 The crowd simulation is validated
using public data from the pedestrian footfall index in Dublin city centre
provided by Dublin’s open data portal7. The model can evaluate the
accuracy of the predictions using unseen data, which it can then use to
update the model creating a continuous learning cycle.

5. Results

5.1. Skyline simulation

A 3D model provides a low barrier to entry for citizens to become

engaged in city planning decisions as no detailed technical knowledge is
needed. The buildings are simulated in 3D and citizens can report if
there are any problem. Fig. 4 shows a comparison between a current

8
https://www.scss.tcd.ie/~whiteg5. Fig. 5. Planned buildings.

6
G. White et al. Cities 110 (2021) 103064

access to what their city will look like in the future as these buildings can
take years to complete.
Fig. 5 shows a proposed building addition, based on the Trans­
america Pyramid that would change the skyline in Dublin. Fig. 5a shows
the current model of the city and Fig. 5b shows the proposed building
plan. This provides easy access to city officials as well as the public to
walk around the city and identify some of the problems that a large
building like this may have in the city, such blocking sunlight or cell
towers. This allows any complains to be heard before the building has
begun construction.
Fig. 6 is the feedback that has been collected from 30 citizens in the
area though an online form. The citizens were asked whether they
approve of the new building proposed in Fig. 5b. From the data collected
78% of citizens approved of the new building, while 22% disapproved.
Citizens were then able to leave comments on why they chose to approve
or disapprove the design of the new building. The comments for people
who disapproved of the building were that “the design didn’t fit in with
Dublin’s architecture” and that “it is too tall and sticks out compared to
surrounding buildings”. This can provide useful insight to urban plan­
ners and policy makers to update the design of the building.

5.2. Green space simulation

The Digital Twin concept can also be used for the creation of green
space throughout the city, this can be by the creation of new park spaces
in the city or planting additional trees. Cities can benefit hugely from the
creation of green spaces with new parks in between buildings. Fig. 7a
shows a simple demonstration of a park that has been created in the city
centre as a place for residents and workers to relax. The amount of cit­
Fig. 7. Green space.
izens using the park can be tracked using sensors in the smart city. As the
digital twin is available online users can give feedback by interacting
with the model and leaving suggestions. This can be the placement of
new benches, trees, plants or outdoor park fitness equipment. The sen­
sors from the smart city can be used to track engagement with the
additional equipment.
Fig. 7b shows a smaller green space created in the city by planting
some additional trees. Simulations using data from the smart city can be
used to ensure that the trees have access to a suitable amount of sunlight,
temperature and water throughout the year. Citizens in the local area
could also give feedback by selecting from a short-list of trees chosen to
be suitable in this area.
Fig. 8 shows the results of the feedback on additional items that
should be included in the new park. Three of the options: more benches,
more summer flowers and park gym equipment were proposed in the
form. The other extruded option for a children’s play pen was suggested
by a user in the form and then added as an option, which become the
second most popular option. This encourages citizen engagement as they Fig. 8. Green space simulation feedback.

can see that the options are being changed based on their feedback. The
most popular option was for park gym equipment, this could be due to
the location of the park in a financial district where employees would
like the option of getting some free exercise outdoors.

5.3. User tagging simulation

The availability of a 3D model allows users to interact with and

report objects in the scene. Fig. 9 shows a user interacting with a street
light in the 3D model. Interacting with the street light gives the user the
option to report an issue such as the light not working or the light
switching on while it is still bright outside. This report can then be sent
to the city council to fix the problem, with a detailed report of the exact
location and problem that has to be fixed. Citizens interact with a
number of different objects in the environment reporting problems such
as litter, antisocial behaviour, traffic congestion, graffiti and mistakes in
Fig. 6. Skyline simulation feedback. the digital twin that need to be re-scanned. This allows for the easy

7
G. White et al. Cities 110 (2021) 103064

Fig. 9. User reporting problem.

generation of additional smart city data that can be used to identify the
problems that are most often reported by citizens.
Fig. 10 shows the user tagging feedback that was collected in the
simulation. We can see in the top left corner of the map there have been
a series of user tags to report problems with litter in this area. This can
then be reported to the Dublin litter wardens and environment health
officers, with the specific location of where the problem is occurring.
Preventative measures, such as increased CCTV or litter warden
deployment can then be implemented in that area to tackle the problem.
Some of the other areas where problems were tagged in the model can
also be seen in the map. These were for problems, such as lights not
working and potholes in the road. Both of these problems can also be
forwarded to the relevant department with the exact location of the Fig. 11. Flooding simulation.
problem. The data can be stored long term to conduct a detailed analysis
over time of how council resources are being deployed to different areas then be evaluated at times when flooding occurs. Additional complexity
and what are the specific problems or locations that are often being can be added to the model to include currents, materials in the river bed,
tagged. sewage and drainage systems. The goal is not to create a perfect model
first time, but to create an open model that can be incrementally
5.4. Flooding simulation improved with additional data and combined with other digital twin
simulations, such as crowd simulations.
Digital Twins can also be very useful for emergency situation simu­
lations as they allow for the simulation of events that happen very 5.5. Crowd simulation
rarely. Fig. 11 shows the simulation of a flooding scenario in the
Docklands. The simulations are able to show how a rise in water level The crowd simulations create two types of pedestrians in the Unity
from the River Liffey would spill into the surrounding streets. From model: a standard Adult agent and an Elderly agent. The average wait
Fig. 11a–f we can see how the increase in water level leads to more times, distance travelled and time in simulation were compared between
damage and how each area in the city centre would be damaged. This the two types. Fig. 12 shows the average wait time and Fig. 13 shows the
information could be used for the effective placement of sand bags and average distance. The average wait time for the adults and elderly in
other flooding countermeasures to protect the areas most at risk. Fig. 12 does not vary much as the percentage of elderly in the simulation
This flooding information can also be used for effective urban increases. However, there is a noticeable difference between the two
evacuation. Urban authorities can make projections about how far the agent types with the elderly agents having a much longer waiting time.
water will rise given the forecasted rain (Alvisi et al., 2006). This will
allow them to identify the areas that are most at risk and begin evacu­
ating people in those affected areas first. Urban authorities will also have
time estimates to evacuate these people as they can forecast how long it
will take for the water level to rise that amount. The flooding model can

Fig. 12. Average wait time of the agents vs. the percentage of the flock that is
Fig. 10. User tagging feedback. of the Elderly type.

8
G. White et al. Cities 110 (2021) 103064

6.2. Infrastructure

The rail bridge near Samuel Beckett bridge is shown as seen in the
physical world and in the digital twin in Fig. 15. Fig. 15a and Fig. 15c
show the comparison of the physical and virtual left side of the bridge.
The virtual representation of the pedestrian path is narrower than the
physical model. The bridge also has legs sticking out that cover the
pedestrian footpath area in the model. Changes to the size of the foot­
path and the position of the bridge were implemented in Unity to make
the simulations more realistic.
Fig. 15b shows the right side of the bridge in the physical world and
Fig. 15d shows how this is represented in the digital twin. The digital
twin is missing the small bridge as seen in Fig. 15b to allow pedestrian
access on this side of the bridge. A small bridge structure as shown in
Fig. 15d is created in Unity to allow for more accurate pedestrian
simulations.

Fig. 13. Average distance travelled by the agents vs. the percentage of the flock
6.3. Mobility
that is of the Elderly type.

There are some urban mobility systems that are not included in the
This is due to the elderly agents having less mobility and movement model. For example, Dublin has a light rail system called Luas that is not
speed, which means that they can take longer to move around obstacles included in the model. The addition of this light rail system would allow
and go through traffic lights. Further experimentation can evaluate how for more detailed urban mobility simulations and the comparison be­
individual street furniture (e.g., bins, lights, footpath width etc.,) effects tween a range of urban mobility systems. There is also a lack of bus stops
the waiting time of pedestrian types as they move through the city. The and road markings in the model, but this information can be added to
results from these simulations can then be used to inform urban planning the model using OpenStreetMap.
and policy.
Fig. 13 shows the average distance that is travelled by the agent
6.4. Smart city
types. The elderly agents distance travelled remains quite constant.
However, the adult distance travelled increases as the percentage of
Data produced from the smart city is what drives the accurate sim­
elderly agents in the simulation increases. This indicates that the adult
ulations in the digital twin. Dublin makes a lot of data available through
agents can use their increased speed to move around any elderly agents
the Dublinked9 open data source. This provides a number of data sources
in the simulation or obstacles. This keeps their average waiting time low
about population, transportation and infrastructure as well as informa­
as shown in Fig. 12, even though the distance they travel increases with
tion about the environment and energy. However, there is a lack of fine
the percentage of elderly people.
grain data, which is often aggregated. This can lead to having to inter­
polate datasets, which leads to less accurate simulations.
6. Limitations of current digital twin

In this section we identify some of the limitations of the current

Digital Twin through the layers of the model outlined in Fig. 1. This is
the first version of the model to be released and these limitations can be
addressed in future versions.

6.1. Buildings

The Docklands model includes a large area of Dublin, some of the

modern buildings along the waterfront are captured in great detail.
However, there are some areas in the model, especially of older build­
ings and open spaces, such as the Bus Depot in Fig. 14 that are repre­
sented by grey boxes. Although the function of the grey box cannot be
determined from the model it does not affect the functionality of the
citizen feedback or crowd simulations.

Fig. 15. Rail bridge that has pedestrian pathways to either side of it, but cannot
be passed in the 3D model.

9
Fig. 14. Physical and virtual bus depot comparison. https://data.smartdublin.ie/.

9
G. White et al. Cities 110 (2021) 103064

7. Conclusion and future work References

The results section has illustrated a range of simulations that can be Abdel-Basset, M., & Mohamed, M. (2018). The role of single valued neutrosophic sets and
rough sets in smart city: Imperfect and incomplete information systems.
carried out by using a digital twin of a smart city. The results from these Measurement, 124, 47–55.
simulations can feed back through the layers of the digital twin model, Abella, A., Ortiz-De-Urbina-Criado, M., & De-Pablos-Heredero, C. (2017). A model for the
as shown in Fig. 1 to advise on real changes in a smart city. The digital analysis of data-driven innovation and value generation in smart cities’ ecosystems.
Cities, 64, 47–53.
twin can be used to engage citizens and get a lot of valuable feedback on Alam, K. M., & El Saddik, A. (2017). C2ps: A digital twin architecture reference model for
key urban planning and policy decisions. The results of skyline and the cloud-based cyber-physical systems. IEEE Access, 5, 2050–2062.
green space planning simulations have shown how user feedback and Albino, V., Berardi, U., & Dangelico, R. M. (2013). Smart cities–definitions, dimensions,
and performance. Proceedings IFKAD, 1723–1738.
suggestions can be collected to provide additional data to inform urban Alvisi, S., Mascellani, G., Franchini, M., & Bardossy, A. (2006). Water level forecasting
planning and policy decisions to allow for additional changes before a through fuzzy logic and artificial neural network approaches. Hydrology and Earth
final decision is made. In the green space planning simulation options System Sciences Discussions, 10(1), 1–17.
Angelidou, M. (2015). Smart cities: A conjuncture of four forces. Cities, 47, 95–106.
can be left open to allow citizens to propose their own ideas, such as a
Anguluri, R., & Narayanan, P. (2017). Role of green space in urban planning: Outlook
children’s play pen in the park. These initial form-based methods can be towards smart cities. Urban Forestry & Urban Greening, 25, 58–65.
extended to a tagging model as shown in the user tagging simulation. Bhavnani, S. P., & Sitapati, A. M. (2019). Virtual care 2.0—a vision for the future of data-
This allows citizens to interact with all the objects in the digital twin and driven technology-enabled healthcare. Current treatment options in cardiovascular
medicine, 21(5), 21.
tag problems or suggestions. These problems or suggestions can then be Boschert, S., & Rosen, R. (2016). Digital twin the simulation aspect. In Mechatronic
sent to the relevant government department with the exact location of futures (pp. 59–74). Springer.
the problem. Bruynseels, K., Santoni de Sio, F., & van den Hoven, J. (2018). Digital twins in health
care: Ethical implications of an emerging engineering paradigm. Frontiers in Genetics,
A digital twin can also be used for a range of other urban planning 9, 31.
decisions and policy. As the model is 3D, it allows for the simulation of Caragliu, A., Del Bo, C., & Nijkamp, P. (2011). Smart cities in europe. Journal of Urban
rare events that require 3D data, such as flooding in the city. This can Technology, 18(2), 65–82.
Caragliu, A., & Del Bo, C. F. (2019). Smart innovative cities: The impact of smart city
then inform the city’s policy of what areas of the city to evacuate first policies on urban innovation. Technological Forecasting and Social Change, 142,
and where to best place sandbags. The crowd simulation in Section 5.5 373–383.
has shown how multiple agents types, such as adults and elderly can be Castelli, G., Cesta, A., Diez, M., Padula, M., Ravazzani, P., Rinaldi, G., … Campana, E. F.
(2019). Urban intelligence: A modular, fully integrated, and evolving model for cities
created in Unity using mobility data from previous experimental studies digital twinning, in: 2019 IEEE 16th International Conference on Smart Cities: Improving
(Oxley et al., 2005). These simulations can be extended to evaluate Quality of Life Using ICT IoT and AI (HONET-ICT) (pp. 033–037).
future urban planning decisions in a smart city such as the impact that a Chin, J., Callaghan, V., & Lam, I. (2017). Understanding and personalising smart city
services using machine learning, the internet-of-things and big data. In 2017 IEEE
change in gradient of a footpath or adding additional street furniture
26th international symposium on industrial electronics (ISIE), IEEE (pp. 2050–2055).
will have on different pedestrians, such as adults and elderly. In future Chiodi, S. I. (2016). Crime prevention through urban design and planning in the smart
work, we plan to extend the public data that we use for the simulations city era: The challenge of disseminating cp-udp in Italy: Learning from europe.
with data from additional IoT services in the surrounding area. These Journal of Place Management and Development, 9(2), 137–152.
Cocchia, A. (2014). Smart and digital city: A systematic literature review. In Smart city
services may be provided by the city council or from private citizens or (pp. 13–43). Springer.
companies in the area. The increased amount of data would allow for R. P. Dameri, et al., Smart city implementation, Progress in IS; Springer: Genoa, Italy.
more realistic simulations with real time information about crowds, B. Dixon, The technology behind the Thailand cave rescue (2018). URL https://www.es
ri.com/about/newsroom/blog/technology-behind-thailand-cave-rescue/.
noise pollution and traffic in the area. This digital twin model with El Saddik, A. (2018). Digital twins: The convergence of multimedia technologies. IEEE
mobility data can be adapted to multiple other scenarios that were not Multimedia, 25(2), 87–92.
investigated in this paper. In future work, we plan to evaluate a concert M. Enzer, A. Bolton, C. Boulton, D. Byles, A. Cook, L. Dobbs, P. A. El Hajj, E. Keaney, A.
Kemp, C. Makri, et al., Roadmap for delivering the information management
simulation that would see an influx of pedestrian and vehicle traffic framework for the built environment doi:10.17863/CAM.38227.
attempting to reach the Three Arena in Dublin (capacity 13,000). Sim­ Glaessgen, E., & Stargel, D. (1818). The digital twin paradigm for future nasa and us air force
ulations could investigate the placement of fire evacuation points as well vehicles, in: 53rd AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics and
materials conference 20th AIAA/ASME/AHS adaptive structures conference 14th AIAA,
as the effect that increasing the capacity of the stadium would have on p.
the traffic and fire evacuation policy. Grieves, M., & Vickers, J. (2017). Digital twin: Mitigating unpredictable, undesirable
emergent behavior in complex systems. In Transdisciplinary perspectives on complex
systems (pp. 85–113). Springer.
CRediT authorship contribution statement
Hashimoto, K., Yamada, K., Tabata, K., Oda, M., Suganuma, T., Rahim, A., Vlacheas, P.,
Stavroulaki, V., Kelaidonis, D., & Georgakopoulos, A. (2015). ikaas data modeling: A
Gary: Conceptualisation, Writing – Original Draft, Software, data model for community services and environment monitoring in smart city. In
Visualisation. 2015 IEEE international conference on autonomic computing, IEEE (pp. 301–306).
Hsu, R. C., Liu, C.-T., & Wang, H.-L. (2014). A reinforcement learning-based tod
Anna: Software, Writing – Original Draft. provisioning dynamic power management for sustainable operation of energy
Lara: Writing – Review & Editing. harvesting wireless sensor node. IEEE Transactions on Emerging Topics in Computing, 2
Siobhán: Writing – Review & Editing, Funding Acquisition. (2), 181–191.
Jalali, R., El-Khatib, K., & McGregor, C. (2015). Smart city architecture for community
level services through the internet of things. In 2015 18th international conference on
Declaration of Competing Interest intelligence in next generation networks, IEEE (pp. 108–113).
I. M. Johnson, Smart cities, smart libraries, and smart librarians.
Kaur, M. J., Mishra, V. P., & Maheshwari, P. (2020). The convergence of digital twin, iot,
The authors declare that they have no known competing financial and machine learning: Transforming data into action. In Digital twin technologies and
interests or personal relationships that could have appeared to influence smart cities (pp. 3–17). Springer.
the work reported in this paper. Lee, A. C. K., Jordan, H. C., & Horsley, J. (2015). Value of urban green spaces in
promoting healthy living and wellbeing: Prospects for planning. Risk management
and healthcare policy, 8, 131.
Acknowledgment Lopez, P. A., Behrisch, M., Bieker-Walz, L., Erdmann, J., Flötteröd, Y.-P., Hilbrich, R., …
WieBner, E. (2018). Microscopic traffic simulation using sumo. In 2018 21st
international conference on intelligent transportation systems (ITSC), IEEE (pp.
This work was funded by Science Foundation Ireland (SFI) under
2575–2582).
grant 13/IA/1885. Luo, W., Hu, T., Zhang, C., & Wei, Y. (2019). Digital twin for cnc machine tool: Modeling
and using strategy. Journal of Ambient Intelligence and Humanized Computing, 10(3),
1129–1140.
Mannion, P., Duggan, J., & Howley, E. (2016). An experimental review of reinforcement
learning algorithms for adaptive traffic signal control. In Autonomic road transport
support systems (pp. 47–66). Springer.

10
G. White et al. Cities 110 (2021) 103064

Medvedev, A., Fedchenkov, P., Zaslavsky, A., Anagnostopoulos, T., & Khoruzhnikov, S. traffic modelling techniques. In Simulating urban traffic scenarios (pp. 115–126).
(2015). Waste management as an iot-enabled service in smart cities. In Internet of Springer.
things, smart spaces, and next generation networks and systems (pp. 104–115). Springer. Q. Qi, F. Tao, T. Hu, N. Anwer, A. Liu, Y. Wei, L. Wang, A. Nee, Enabling technologies
Menouar, H., Guvenc, I., Akkaya, K., Uluagac, A. S., Kadri, A., & Tuncer, A. (2017). Uav- and tools for digital twin, Journal of Manufacturing Systems doi:https://doi.org/10
enabled intelligent transportation systems for the smart city: Applications and .1016/j.jmsy.2019.10.001. URL http://www.sciencedirect.com/science/article/pii/
challenges. IEEE Communications Magazine, 55(3), 22–28. S027861251930086X.
Mohammadi, M., & Al-Fuqaha, A. (2018). Enabling cognitive smart cities using big data Rosen, R., Von Wichert, G., Lo, G., & Bettenhausen, K. D. (2015). About the importance
and machine learning: Approaches and challenges. IEEE Communications Magazine, of autonomy and digital twins for the future of manufacturing. IFAC-PapersOnLine,
56(2), 94–101. 48(3), 567–572.
Mohammadi, N., & Taylor, J. (2019). Devising a game theoretic approach to enable Sánchez, L., Elicegui, I., Cuesta, J., Muñoz, L., & Lanza, J. (2013). Integration of utilities
smart city digital twin analytics. In Proceedings of the 52nd Hawaii international infrastructures in a future internet enabled smart city framework. Sensors, 13(11),
conference on system sciences. 14438–14465.
Mohammadi, N., & Taylor, J. E. (2017). Smart city digital twins. In 2017 IEEE symposium Soon, K. H., & Khoo, V. H. S. (2017). Citygml modelling for singapore 3d national
series on computational intelligence (SSCI), IEEE (pp. 1–5). mapping, ISPRS - International archives of the photogrammetry. Remote Sensing and
Neirotti, P., De Marco, A., Cagliano, A. C., Mangano, G., & Scorrano, F. (2014). Current Spatial Information Sciences, 37–42. XLII-4/W7 https://www.int-arch-photogramm-r
trends in smart city initiatives: Some stylised facts. Cities, 38, 25–36. emote-sens-spatial-inf-sci.net/XLII-4-W7/37/2017/.
Oldenbroek, V., Verhoef, L. A., & Van Wijk, A. J. (2017). Fuel cell electric vehicle as a Tao, F., Cheng, J., Qi, Q., Zhang, M., Zhang, H., & Sui, F. (2018). Digital twin-driven
power plant: Fully renewable integrated transport and energy system design and product design, manufacturing and service with big data. The International Journal of
analysis for smart city areas. International Journal of Hydrogen Energy, 42(12), Advanced Manufacturing Technology, 94(9–12), 3563–3576.
8166–8196. L. Wang, D. Sng, Deep learning algorithms with applications to video analytics for a
Oxley, J. A., Ihsen, E., Fildes, B. N., Charlton, J. L., & Day, R. H. (2005). Crossing roads smart city: A survey, arXiv preprint arXiv:1512.03131.
safely: An experimental study of age differences in gap selection by pedestrians. White, G., & Clarke, S. (2020). Urban intelligence with deep edges. IEEE Access, 8,
Accident Analysis and Prevention, 37(5), 962–971. https://doi.org/10.1016/j. 7518–7530. https://doi.org/10.1109/ACCESS.2020.2963912.
aap.2005.04.017. http://www.sciencedirect.com/science/article/pii/S0001457 White, G., Liang, Z., & Clarke, S. (2019). A quantified-self framework for exploring and
505000795. enhancing personal productivity. In 2019 international conference on content-based
Parra, L., Sendra, S., Lloret, J., & Bosch, I. (2015). Development of a conductivity sensor multimedia indexing (CBMI), IEEE (pp. 1–6).
for monitoring groundwater resources to optimize water management in smart city White, G., Nallur, V., & Clarke, S. (2017). Quality of service approaches in iot: A
environments. Sensors, 15(9), 20990–21015. systematic mapping. Journal of Systems and Software, 132, 186–203.
Pramanik, M. I., Lau, R. Y., Demirkan, H., & Azad, M. A. K. (2017). Smart health: Big data White, G., Palade, A., Cabrera, C., & Clarke, S. (2019). Autoencoders for qos prediction at
enabled health paradigm within smart cities. Expert Systems with Applications, 87, the edge. In 2019 IEEE international conference on pervasive computing and
370–383. communications (pp. 1–9). PerCom. https://doi.org/10.1109/
D. Puri, The vital role of technology in the thai cave rescue mission (2018). URL https PERCOM.2019.8767397.
://www.networkworld.com/article/3291787/the-vital-role-of-technology-in-the-th Williamson, B. (2015). Educating the smart city: Schooling smart citizens through
ai-cave-rescue-mission.html. computational urbanism. Big Data & Society, 2(2), 2053951715617783.
Pyatkova, K., Chen, A. S., Djordjević, S., Butler, D., Vojinović, Z., Abebe, Y. A., & Zhao, L., Wang, J., Liu, J., & Kato, N. (2019). Routing for crowd management in smart
Hammond, M. (2019). Flood impacts on road transportation using microscopic cities: A deep reinforcement learning perspective. IEEE Communications Magazine, 57
(4), 88–93.

11