Simulating the re-concentration of dispersed habitation in the peripheral urban area using an artificial society approach

ABSTRACT

This study proposes an exploratory urban morphology agent-based model (ABM) to simulate the combined impact of the establishment of hub facilities that urban residents can conveniently access, policies for promoting human interactions around them, and the introduction of trams amidst dispersed habitations in peripheral urban areas. The proposed model offers a new perspective on ways to improve the urban environment. It describes a shift from dispersed habitation to concentrated habitation in a bottom-up manner through behavioural changes at the micro level, which lead to productive human interactions in an urban setting. Specifically, the model experimentally demonstrates a trade-off between increased human interactions caused by the introduction of hub facilities that attract a diverse range of activities, and policies that promote such interactions and development amongst dispersed habitation. Additionally, the model suggests that the direction of urban growth is a consequence of collective action, implying that collaborative efforts can facilitate its improvement.

KEYWORDS:

• Urban morphology
• agent-based simulation
• land use changes
• smart growth
• transit oriented development

1. Introduction

1.1. Measures for dispersed habitation accompanying urban expansion

Global population has exponentially increased throughout the twentieth and twenty-first centuries. Consequently, land – a finite resource – has undergone rapid development worldwide, which has adverse environmental impacts. Urbanisation, in particular, is expanding at a rate far higher than that of population growth, and more than half of the global population lives in cities (Angel et al. 2011; United Nations, Department of Economic and Social Affairs 2012; Gerland et al. 2014; Bacci 2017).

The benefits of cities – a concentration of diverse complementary multiple activities, people, and cultures in one place – cannot be overemphasized. The so-called agglomeration economies in cities have two aspects. First, localisation economies indicate geographic concentrations of inter-connected individuals and groups in a particular field. Second, urbanisation economies, also known as urban agglomerations, emerge when dissimilar individuals and groups are located near each other (Krugman 1996; Irwin 2010; Levin et al. 2013; Beeson 2017). It is not only commercial organisations that benefit from such sharing, matching, and learning (Duranton and Puga 2004); households, too, can avail of social contacts, the quality of the neighbourhood, or the proximity to schools and shops (Elkin et al. 1991; Behan et al. 2008; Wey 2019; Carmona 2021).

In addition to covering the soil, urbanisation places additional load on the land – for example, through the development of lifelines such as water, sewerage, and electricity grids and transport networks. Therefore, urbanisation exerts a greater negative impact on natural and social environments than any other form of land use (Batisani and Yarnal 2009; van Delden et al. 2016). Consequently, dealing with growing urbanisation has become a major global concern.

Recently, guidelines for managing land use have been collaboratively developed by researchers from various disciplines across Europe. Addressing the three aspects of sustainability – the economy, ecosystem, and society – they compared and evaluated the following Scenarios of typical urbanisation as consequences of interventionist urbanisation policies (Cotella et al. 2020):

• Compact urbanisation: often a consequence of policies that address urbanisation in terms of revitalisation, enrichment, and redevelopment of urban areas.

• Polycentric urbanisation: often a consequence of policies that straddle space and time, such as collaboration between public transport and urban development.

• Diffuse urbanisation: often a consequence of liberal policies, including the promotion of privately owned vehicles and home ownership, which is the opposite of bureaucratic planning.

Among the urban forms resulting from these Scenarios, the first and second are largely positively evaluated. However, there are several related concerns such as over risk management and the potential imbalance with other cities in compact urbanisation, and access and traffic noise in polycentric urbanisation. The third urban form, by contrast, is negatively evaluated in almost every aspect, including high transport costs, low energy efficiency, environmental pollution, negative influences on biodiversity, and significant social separation. Its only strength is the convenience it offers to individuals.

Diffuse urban form is one of the significant characteristics of urban sprawl. There is vast anti-sprawl literature designed for both the public and academics. Despite some agreement about the major forms of sprawl and their disadvantages, the definition of ‘sprawl’ remains ambiguous. Although there is insufficient evidence that strategies for containing urban sprawl or densification are effective and that their social benefits outweigh their negative side-effects, political support for those strategies is substantially expanding (Millward 2006; Behan et al. 2008; Rubiera-Morollón and Garrido-Yserte 2020).

In such a context, what is an effective approach for dealing with complex issues such as sustainable urban growth? Land use, including urbanisation, is a consequence of the interaction of various mutually influencing factors. Additionally, these elements are not limited to authorities with legal rights and political or economic power. For example, decisions may be made because stakeholders, including private developers, predict and respond to the decisions of such authorities. A major characteristic of urbanisation is that it ‘emerges’ from numerous individual and collective decisions made by residents or groups of various sizes, regarding where and how to stay, work, study, and enjoy interactions, considering constraints such as affordability and mobility (Batty 2007; Frenkel and Ashkenazi 2008; Webb et al. 2018).

A common method of conventional urban planning is to present an ideal and restrict the activities of individuals and groups in the process of realising it. However, in democracies, such a top-down exercise of power is difficult to impose (Sieverts 2003; Healey 2017). Thus, urban planning requires a multi-dimensional, multi-sector, and multi-stakeholder approach.

Drawing on these concerns, this study presents an approach to improve dispersed habitation by generating a bottom-up ‘emergence’ created by appropriate interventions in resident behaviour by policymakers/implementors.

1.2. Transport oriented development

Among the attempts made worldwide to realise urban sustainability, we cannot overlook the concept of Transport Oriented Development (TOD). Since TOD was developed by Peter Calthorpe in the early 1990s as an alternative to low-density and huge-range residential suburban developments, its basic principles and policies have been effective for at least three decades (Calthorpe 1993). TOD commonly refers to the integration of transport and land use in the form of compact urban development at public transport nodes with high-quality walkable environment (Calthorpe 1993; Ewing et al. 2017; Zhang et al. 2019b; Hrelja et al. 2020; Su et al. 2021). This concept was inspired by the Garden City ideal posited by Ebenezer Howard (Howard 1902), in terms of relatively high-density and small clusters of land use and connection through transport networks (Caves 2004). Originally, TOD was understood as planning principles based on universal models and definitions (Calthorpe 1993; Cervero and Kockelman 1997; Rogers 1997). For example, the three Ds principle – high density, diversity of land-use patterns, and habitable design – used to be the focus (Cervero and Kockelman 1997). Thereafter, as it was applied more widely, a diversity of more flexible approaches and definitions for TOD were accepted. Representatively, another three Ds – accessibility to destination, distance to transit, and demand management – were added (Cervero 2004; Ewing and Cervero 2010; Curtis 2012). Additionally, TOD is considered an effective solution to the serious and growing problems of climate change and global energy security, as it can create dense and walkable communities that significantly reduce the need for driving (US High Speed Rail Association 2015; Hrelja et al. 2020; Ibraeva et al. 2020). Against this background, TOD, as an integrated planning strategy towards urban sustainable growth, has received increasing attention from researchers and practitioners, and gained increasing popularity in many cities worldwide (Thomas et al. 2018; Hrelja et al. 2020; Ibraeva et al. 2020). However, there has been a lack of attention to apply TOD in non-metropolitan, low-density, and provincial contexts (Nigro et al. 2019; Hrelja et al. 2020). In this context, this study discusses the potential of applying TOD beyond the street scale to the regional city scale, smaller than the metropolitan scale, to counteract dispersed habitation.

1.3. AIM of the research

Using a simulation experiment, this study examined interventionist policies designed to achieve sustainable urban growth, focussing on changes in spatial behaviour in the form of dispersed habitation in peripheral urban areas, as a consequence of collective behaviour of residents. The simulation is based on the research findings of the agent-based, operational land use-transport interaction (LUTI) model, which is an effective tool to examine the emergence of bottom-up urban forms based on collective behaviours. Thereafter, the study expanded the model by simultaneously introducing the benefits of human interaction at the micro-level in cities and the given condition of dispersed habitation. The model was further developed by considering the introduction of the tram – a transport policy that not only represents an alternative to privately owned vehicles but also has the potential to exert an influence on face-to-face interaction. This artificial society model allowed us to discuss an approach for sustainable urban growth that evaluates factors that could improve dispersed habitation. It is based on promoting human interaction and transport policies in cities.

This study did not aim to improve the urban environment in a particular location. Rather, it intended to gain insights into changes in urban forms caused by individual residents’ behaviour. Specifically, it sought to answer the following questions:

• How can we model improvements in dispersed habitation in peripheral urban areas, which is an obstacle for sustainable urban growth?

• What type of resident behaviour would lead to improvement?

• What are the factors that would induce such behaviour?

The model suggests that the direction of urban growth is a consequence of collective action, implying that collaborative efforts can improve it (Cotella et al. 2020).

1.4. Structure of the article

The remainder of this article is organised as follows. Section 2 reviews related studies to explain the scientific basis of the study’s urban morphological agent-based model (ABM) and describes the position of the model. Section 3 discusses the basic constitutive elements of the model, based on the Overview, Design concepts, and Details (ODD) protocols. Sections 4 to 6 report a simulation experiment, based on several Scenarios of interventionist policies for urban expansion, and discuss the results. They also present test results on the model’s validity by comparing the model’s output with insights gained from real cities. Finally, Section 7 summarises the major findings of the model and concludes the study by discussing the possibilities for further development.

2. An overview of preceding studies and position of the current study

2.1. Agent-based land use-transport interaction model

Changes in land use, including urbanisation, stem from various driving forces (e.g. social, economic, political, and physical factors) and complex interactions among them (Gimblett et al. 2001; Mitsuda and Ito 2011; Meyfroidt et al. 2018). The basic principle, that land use influences travel behaviour, and vice versa, has been acknowledged by many researchers and is supported by empirical findings (Batty et al. 1999; Acheampong and Silva 2015; Kii et al. 2016). Moreover, interdisciplinary research has examined the laws that govern the relationship between urban travel behaviour and land use over the past half a century. However, in urban planning, there is a strong inclination to draw from experience because experiments in actual cities are nearly impossible.

Accordingly, considerable efforts have been made to develop models and simulations of changes in traffic and land use to explore urban growth strategies. They have led to the development of the land use-transport interaction (LUTI) model, which considers the interaction between land use by households and firms and transport as a system to support decision-making (Batisani and Yarnal 2009; Acheampong and Silva 2015; Ding 2019). Additionally, a concept that considers artificial systems such as cities as a form of self-organisation based on collective interaction among many autonomous factors and their adoptive behaviour has been attracting scholarly attention (Epstein and Axtell 1996; Batty 2007; Kwapień and Drożdz˙2012; Rauws 2016). Drawing on complexity sciences, the agent-based model (ABM) is a unique modelling approach that posits that each agent learns their behaviour through interaction with the given environment, and can revise and improve it (Axtell and Epstein 1994; Epstein 2006; Gilbert 2008; Railsback and Grimm 2011; Shook and Wang 2015). The ABM has enabled us to simulate the bottom-up emergence of macro-level social phenomena, based on individual behaviour.

The application of the ABM to the modelling of land use began with the classical cellular automaton (CA), as employed in Schelling’s model of segregation, in which the condition of each cell represented a specific type of land use. This abstract model of urban morphology describes the emergence of simple patterns (Schelling 1971). Since then, owing to the development of a range of urban ABMs, complex land-use patterns in cities at the macro-level – including the emergence of cities and sprawl – have been expressed as forms of self-organisation employed by agents such as households and firms. These urban ABMs have long been developed and operated for studies on a range of urban issues. The SLEUTH (Slope, Land-cover, Exclusion, Urbanisation, Transportation, Hillshade) is an urban growth model based on CA, mainly used to analyse urban sprawl and explore future Scenarios of land-use changes and urban growth. It has been more application-orientated and calibrated with real-world data in the process of application to various city regions (Clarke et al. 1997; Silva and Clarke 2002; Rafiee et al. 2009; Zhou et al. 2019). The DUEM (Dynamic Urban Evolutionary Model) is an urban growth simulation software consistent with GIS data inputs, outputs, and related functionality and oriented to the analysis of spatial dynamics. It can accept a wide range of hypothetical and real applications, and present ways in which land uses can be structured through their life cycles (Batty et al. 1999; Xie and Batty 2005; Feng et al. 2016). The LEAM (Land use Evolution and impact Assessment Modelling) framework has been developed as a comprehensive urban planning support system to assess the expansion of regional scale sprawl and environmental changes (Deal and Schunk 2004; Deal et al. 2013, 2017). The ALMA (Agent based Land Market) is a new bilateral agent-based land market model, which explicitly simulates spatial and market interactions between buyers and sellers based on different decision rules in a homogeneous landscape. It demonstrates that urban expansion is led by changing the relative market power of buyers and sellers through land rents change (Filatova et al. 2009; Michels et al. 2020). Additionally, ABMs have been used for other urban context problems such as simulations of housing land use during population growth and decline (Haase et al. 2010), formation of slums (Patel et al. 2012), evaluation of housing policies for creating compact cities (Ma et al. 2013), and changes in volume of travel, parking requirements, and CO₂ emissions, which might result from the large-scale uptake of shared and self-driving vehicles (Martinez and Viegas 2017).

2.2. Agent-based human interaction model

Jacobs (1961) emphasised that a significant function of cities was to serve as lively and energetic places for people to interact. Oldenburg (1989) argued that a ‘third place’, comprising streets, squares, and public facilities where people could conveniently gather and interact, was essential to making cities vibrant. Recently, several studies have suggested that in cities that are socially tolerant and have concentrations of diverse residents, the creativity produced by strong human interaction fosters economic growth through improvements in productivity and innovation (Florida 2005; Belitski and Desai 2016).

The benefits of interaction are not only economic; cities that become crucibles with diverse, creative populations offer rich cultural and entertainment experiences (Florida et al. 2011; Bertolini 2020). Consequently, more cities are aiming for smart growth and have introduced policies such as mixed land use, increased walkability, and promotion of population concentration (Henderson 2010; Hrelja et al. 2020). The emergence of this type of human interaction has created a framework in which the social and economic activities of the city are larger than the sum of its constitutive elements (i.e., the activities of individuals and groups).

The ABM is useful in exploring this type of urban system because it can simulate complex social phenomena. Accordingly, research on ABMs has focussed on the relationship between urban forms and human interactions. Spencer (2012) examined the influence exerted by a place on the evolution of social networks through human interaction, using an ABM of social dynamics drawn from evolutionary economic geography. Liu and Silva (2013) developed a spatially explicit model of the market dynamics of the location of firms and households, based on the strength of human interaction, to evaluate industrial policies for sustainable development. Malik et al.’s (2015) creative urban model facilitated the detailed examination of the dynamism of the formation of creative clusters by setting agents with multiple attributes. The model suggested the possibility that knowledge spillover through human interaction induces a bottom-up emergence through the pursuit of social fairness.

2.3. Where is this study situated?

Through a simulation using an abstract model based on a regional city in Japan, Nagai and Kurahashi (2021) introduced a new perspective on bottom-up constraints in dispersed habitation in urban peripheries through the benefits of productive human interaction at the micro level, drawing from the two conceptual frameworks of the ABM studies mentioned above. However, considering that many Japanese cities have already expanded, owing to motorisation (Eaton and Eckstein 1997), the effectiveness of a policy of maintaining a concentrated urban structure is doubtful. Additionally, Japan is mountainous, with a relatively small amount of habitable land and a declining population (Koike et al. 2000; Ministry of Land, Infrastructure and Transport 2017). Therefore, it is unrealistic to construct new concentrated cities in Japan.

Against this background, this study contributes to the literature through our dynamic urban ABM, spatially explicit simulation of Scenarios in which the existing dispersed habitation in urban peripheries is dissolved. By integrating the sub-model components of travel behaviour, relocation, and human interaction and by linking input and output, our model simultaneously included the social complexity of human behaviour and urban forms while stylising them. It enabled us to discuss the bottom-up dissolution of dispersed habitation related to changes in urban form at the macro-level caused by the collective behaviour of individual residents at the micro-level.

Furthermore, the introduction of trams is a well-known method of providing an alternative to privately owned vehicles in cities, to improve the urban living environment by reducing traffic jams and air pollution, and saving energy. It is easier to lay tracks for trams than for conventional trains. Trams are being increasingly introduced as a method of transport for short distances, particularly in European cities. They are also being reintroduced in Japan (Oba et al. 2008; Utsunomiya 2016). Additionally, by modelling the introduction of the tram to connect city centres, this study attempted to overcome the limitations of Nagai and Kurahashi (2021), who discussed only the conventional railway network connecting city centres to suburbs.

3. Agent-based model of urban dynamics

The experimental model examined in this study uses the urban dynamics model presented by Nagai and Kurahashi (2021) as its basic framework, and develops it by changing the conditions under which the simulation begins, diversifying the transport means and refining its parameters. Specifically, the artificial society model – spatially and explicitly – expresses the agents as both individuals using the transport network and relocating households. The model aims to abstract actual social phenomena by focussing on changes in urban land use patterns caused by the autonomous behaviour of urban residents. However, it does not consider irregular changes in land prices due to the influence of real-estate market dealings or direct urban policies. Furthermore, to enhance its applicability to areas other than those examined in the study, the model does not consider geographical conditions such as topography and climate, generational differences among households and individuals, and preferences for a particular behaviour. The outline of the experimental model is described based on ODD protocols (Railsback and Grimm 2011).

3.1. Purpose

The ABM simulates the effects of the relationships between the installation of hub facilities for urban residents, policies for promoting resident interaction, and the introduction of trams for dispersed habitation in urban peripheries through the autonomous behaviour of each resident. Additionally, we aim to gain insights into the micro–macro linkage in the urban system in various Scenarios through the simulation.

3.2. Entity and scale of the experimental model

The entity is a two-dimensional urban scheme with active household agents, both of which are spatially explicit. Figure 1 shows the outline of the urban scheme. It abstracts a typical regional city in Japan. The central business district (CBD) and the commuter town in the suburbs along the railway line was originally planned and built according to a zoning arrangement intended to separate residents’ work lives from their personal lives. The commuter town was then transformed through repeated relocations, autonomously executed by households because of motorisation. Consequently, many residences are dispersed in low-density surrounding the CBD (Tsai 2005; Ding et al. 2022). As shown in the figure, there are two stations, a railway line that connects them, and a major road. To simplify the model, traffic control, such as traffic lights at crossings and one-way roads in the urban scheme, was not considered. It was assumed that the roadway and sidewalks were evenly distributed in high-density.

Figure 1. Dispersed habitation urban scheme.

The number of household agents was set at 1,000. As one household agent corresponds to 10 real households, the model simulated the behaviour of 10,000 real households. The same number of workplaces was placed in the CBD, in accordance with the normal distribution around the central station. Additionally, one hub facility (HUB), which would attract diverse activities to the city centre and can be conveniently used by urban residents, was placed in the CBD. Whether the HUB is run by the public or a private sector body or a combination of the two is inconsequential. A radius of 500 m around the HUB was considered in implementing policies to promote interactions.

3.3. State variables of the household agent

The state variables of the household agent were as follows:

• Coordinates of the residence

• Coordinates of the workplace

• Currently selected linked trip

• Value list of the linked trip

As the study focussed on changes in land use patterns through household relocation, it adopted the assumption of an exogenous workplace. The coordinates of the workplace corresponding to a certain household agent were always fixed. A linked trip is a series of travel behaviours, beginning at the starting point till the arrival at the destination. An unlinked trip comprises each method of transport serving as a constitutive element of a linked trip.

3.4. Process overview and scheduling

The household agents engage with the corresponding travel behaviour to commute to work or place of study, based on the value list of the linked trip in the urban scheme. Following the learning period, in which the process is repeated 30 times, the transport methods converge into a single method. Subsequently, 10 candidates for relocation, which are within, or not very far from, the distribution of the residences as a whole, are proposed to a few randomly selected household agents. They relocate to the candidate with the lowest total cost of living. The total cost of living is calculated on the basis of two factors: travel cost and rent. These relocations change the land use pattern.

In a single simulation round, the relocation cycle is repeated 20 times. It is assumed that a single relocation cycle in the simulation (30 rounds of travel behaviour in the model) corresponds to two years in the real world. Thus, a single round of simulation corresponds to 40 years in the real world. It implies that household agents would relocate twice, on average, over 40 years. As relocation tends to occur amid a change in family structure, such as a birth or separation (Kulu and Milewski 2007; Yamada and Deguchi 2016), this scheduling is sufficiently valid.

3.5. Sub-model of travel behaviour

The household agent’s travel behaviour was spatially and explicitly processed on the urban scheme to reflect the empirical finding that urban variables such as the travel distance, the available means of transport, and transport density influence travel behaviour in the model. Each travel behaviour of a household agent followed the linked trip it selected.

Six types of linked trips were selected for the model, as shown in Figure 2. The most main means of transport among the unlinked trips, which constitute a linked trip, is referred to as the ‘representative means of transport’ of the linked trip. The representative means of transport for most household agents were privately owned vehicles, reflecting the advanced state of motorisation in the real world at the start of the simulation. Each agent started from their residence. Once all the household agents reached their workplace, they began leaving for the HUB. After spending some time there, they went back to their residence. Once they returned to their residence, the total travel cost, $C$, was calculated with the following formula:

Figure 2. Six types of linked trips.

(1) $C=w_r{C}_r+w_c{C}_c+w_f{C}_f-w_{P}P$(1)

where $C_r$, $C_c$, $C_f$, and $P$ represent time, fees, fatigue, and interaction values, respectively; and $w_r$, $w_c$, $w_f$, and $w_P$ represent each agent’s preference bias. The study emphasised homogeneity rather than differences among household agents; thus, it was assumed that all household agents had equal preference bias. The $\epsilon$-greedy method – a method for reinforcement learning – was used by household agents to explore transport methods, reflecting the finding that individuals do not select transport methods and routes rationally, based on complete information, but are often bound by habitual rationality (Fujii 2008). Based on the cost, the value of the linked trip, selected as the $i$-th trip from the value list, $V_i$, was updated according to the following formula:

(2) $V_i\leftarrow \alpha (-C)+(1-\alpha )V_i$(2)

The next travel behaviour followed the linked trip selected from the value list. The household agents converged the linked trips to a single trip after the learning period, in which they repeated their travel behaviour 30 times.

3.5.1. Traffic jams

Traffic jams during morning and evening peak hours are a serious problem worldwide. Traffic jams incur a variety of social costs, including an increase in time loss, financial cost, unpleasantness, fatigue, and air pollution (Behan et al. 2008; Zhang et al. 2019). To reflect a simplified road transport network, the study assumed that traffic jams were generated as a consequence of local interactions among agents. Specifically, a traffic jam is generated when the number of household agents travelling in their privately owned vehicles in the neighbourhood simultaneously as a household agent travelling in a privately owned vehicle exceeds a certain limit. Consequently, the travelling speed of the household agent in question decreases. The reduction in speed due to traffic jams increases time loss, financial costs, and fatigue costs. Additionally, the fluctuation in travelling time, owing to traffic jams, is considered an additional cost that cannot be ignored as it could be a factor for additional risk or uncertainty. Furthermore, its influence becomes more noticeable in proportion to the severity of the traffic jams (Bhat and Sardesai 2006; Mori et al. 2015). Based on these findings, the traffic jam fatigue cost was added to the household agents who were in the middle of a traffic jam.

3.5.2. Interaction value

A 500 m radius around the HUB was considered in the implementation of policies for promoting interaction. This area was under the influence of diverse activities. The distance was set with reference to the concept of ‘Ped Shed’ (Hebbert 2003). When household agents travelling on foot or on a bicycle interacted face-to-face within this area (in other words, when they geographically agglomerated on the model plain), a useful, efficient, and creative cultural or entertaining experience was generated. It was assumed that this benefit was enjoyed as an ‘interaction value’. The urban environment is improved (via innovation, entrepreneurship, and economic productivity) when land use within a city is diversified and interaction density increases (Knudsen et al. 2008; Malik et al. 2015). The study emphasised the positive feedback loop in which many people, staying in a particular place, attracted others and enhanced the strength of activities in that place (Kakoi et al. 2010). Accordingly, the interaction value, $P$, was measured using the formula given below:

(3) $P={\eta }_{ir}{D}_{ir}$(3)

where $D_{ir}$ represents the number of other household agents who travel on foot or on bicycle within the $r_{ir}$ metre radius of the household agent in question. ${\eta }_{ir}$ represents the interaction promotion coefficient. The result of multiplying the interaction value, $P$, with the preference bias, $w_P$, is cancelled out by the total travel cost. The interaction promotion coefficient can be regarded as the degree of collaborative effort, resulting in further interaction, in proportion to the accumulation of pedestrians within the area in question. Empirical findings demonstrate that the characteristics required for an urban environment to feature valuable human interaction include the following (Knudsen et al. 2008; Drennan and Kelly 2010; Florida et al. 2011; Landry 2012; Malik et al. 2015; Belitski and Desai 2016; Bertolini 2020)

• Affordable mobility

• Mixed land/building use

• Social diversity

• Social tolerance

Consequently, the interaction promotion coefficient can be enhanced via collaborative efforts in the area as a whole, such as the development of pleasant walkways and cycle lanes, the expansion of diverse and attractive shops and places, and the hosting of open events that offer mobility, intermingling, diversity, and tolerance. When the coefficient increases, the benefits of mobility with interaction also increase, which further cancels out the total travel costs. Consequently, this coefficient can be regarded as a coefficient of gain. Policies for improving the interaction promotion coefficient are, hereafter, referred to as ‘interaction policies’. An excessive accumulation of pedestrians or cyclists could result in harm, equivalent to a traffic jam caused by automobiles. However, the study did not consider such negative effects as it assumed that the policy of developing pleasant walkways and cycle lanes to promote interaction would ease the congestion.

3.6. Sub-model of relocation

After all the household agents converged the transport methods into a single method, having completed the learning period, we randomly selected 10% of the household agents for relocation. The relocation occurred spatially, explicitly in the urban scheme. The household agents in question were presented with a list of 10 destinations within, or not very far from, the distribution of the residences as a whole. Each agent relocated to the destination with the smallest assumed living cost. In the process, households faced several restrictions in satisfying their preferences (Moeckel 2017). Among them, travel cost (or travel time) and rent were emphasised (Iacono et al. 2008; Hunt 2010; Pagliara et al. 2010; Taniguchi et al. 2010; Acheampong and Silva 2015). Thus, the assumed total living cost of the destination, $C^l$, was calculated as follows:

(4) $C^l=C+R$(4)

where $C$ represents the assumed total travel cost to the destination. The household demanded a high level of accessibility to work and to places of shopping and leisure activities (Iacono et al. 2008). Additionally, travel behaviour before relocation became a habit and influenced the relocation destination and the travel behaviour after relocation (Fujii 2008). Therefore, the total assumed travel cost, $C$, was calculated by measuring the assumed travel behaviour, starting from the destination and based on the transport method on which the household agent converged through learning. $R$ refers to rent of the destination. Owing to improvements in efficiency and productivity caused by accumulation, there is a positive correlation between the degree of development concentration and rent (Drennan and Kelly 2010; Koster and Rouwendal 2012). Therefore, the rent of the destination, $R$, was calculated as follows:

(5) $R={\eta }_R^r{I}_i^r(\frac{A^r}{A})+{\eta }_R^j{I}_i^j(\frac{A^j}{A})$(5)

where $I_i^r$ and $I_i^j$ refer to the number of residences and workplaces, respectively, within the radius, $r_R$, of the destination. $A$ represents the size of the area in question. $A^r$ and $A^j$ represent consumption areas by residence or workplace, respectively. ${\eta }_R^r$ and ${\eta }_R^j$ represent transformation coefficients for the rent of the residence or workplace, respectively. Figure 3 illustrates how the relocation of household agents is processed in the model with reference to the assumed total living cost of the destination, $C^l$ (and the assumed total travel cost $C$ and rent $R$). Thus, the rent of the destination increased in proportion to the accumulation of residences and workplaces. Specifically, local interactions among the households, and between the household and the environment, influenced the land use pattern through changes in rent.

Figure 3. Process of relocation of household agents.

3.7. Initialisation and input

Table 1 lists the parameter settings for the urban scheme and household agents. The setting of these values, particularly the currency conversion coefficients, requires utmost care because setting this coefficient leads to automatic changes in profit evaluation. Consequently, the study carefully set these values, with reference to various sources, including socio-demographic data and other statistics published by official institutions and previous studies while assuming a typical regional city in Japan. The validity of the model, including these values, was supported by the contrast between the simulation results and the real city, based on pattern-oriented modelling (POM).

Table 1. Parameter settings based on socio-demographic data and other statistics published by official institutions (Ministry of Land, Infrastructure and Transport 2017; MIAC 2017) and previous studies (Taniguchi et al. 2010; Nagai and Kurahashi 2021).

Parameters of urban scheme	Setting values
Bicycle parking fare $m_{bp}$	1.5 (EUR)
Railway fare (one way) $m_{rw}$	1.6 (EUR)
Vehicle parking fare $m_{cp}$	8 (EUR)
Gas price $m_g$	0.01 (EUR/h)
Tram fare (one way) $m_{tr}$	1.6 (EUR)
Land rent coefficient of residence ${\eta }_R^r$	1.5 (EUR)
Land rent coefficient of workplace ${\eta }_R^j$	15 (EUR)
Area size of residence $A^r$	140 (m${ }^2$)
Area size of workplace $A^j$	140 (m${ }^2$)
Parameters of household agents	Setting values
Walk velocity $V_w$	4 (km/h)
Bicycle velocity $V_b$	12 (km/h)
Railway velocity $V_{rw}$	55 (km/h)
Vehicle velocity $V_N$	24 (km/h)
Vehicle velocity on major road $V_H$	60 (km/h)
Tram velocity $V_{tr}$	12 (km/h)
Time cost coefficient ${\eta }_t$	0.5 (EUR/min)
Fatigue cost coefficient for walking $f_w$	1.8 (EUR/min)
Fatigue cost coefficient for bicycle $f_b$	4 (EUR/min)
Fatigue cost coefficient for railway $f_{rw}$	0.6 (EUR/min)
Fatigue cost coefficient for vehicle $f_c$	0.6 (EUR/min)
Fatigue cost coefficient for railway $f_{tr}$	0.6 (EUR/min)
Greedy rate $\epsilon$	0.5
Learning rate $\alpha$	0.5
Attenuate rate $\gamma$	0.94

3.8. Observation and evaluation

Changes in urban form were evaluated by observing the results of the simulation of each Scenario with reference to the following indicators:

• Proportion of each representative transport method

• Total CO ${ }_2$ emissions¹ (expressed in percentage using the level at the start of the simulation as the base)

• Average travel time

• Standard deviation of residence distribution (east–west and north–south)

• Distribution map of residences (coloured using the representative transport method)

Each Scenario was executed eight times under the same conditions. The proportion of each representative transport method, total CO ${ }_2$ emissions, average travel time, and standard deviation of the residence distribution are the averages of the execution results. The distribution map of residence was arbitrarily generated from these execution results.

4. Experiment 1: simulation from dispersed habitation

4.1. Conditions for experiment 1

This section reports the simulation experiment. It started in the state where dispersed habitation had already progressed. The lower row in Table 2 shows the value of each indicator at the start of the simulation. The lower figure in Figure 4 denotes the distribution of residences and workplaces at that point. Most residences were dispersed in low-density surrounding the CBD and approximately 90% of the households used privately owned vehicles. It shows that the state, after the relocation cycle, was repeated 20 times (equivalent to 40 years) from the initial state, in which all residences were concentrated around a suburban station, in accordance with the zoning arrangement intended to separate work and life (see the upper row of Table 2 and the upper figure in Figure 4). Under these initial conditions, a simulation experiment was performed on the effect of installing a HUB, which is open to urban residents, and the implementation of interaction policies around it. There are two options for the location of the HUB:

Figure 4. Distribution of residences at the initial state and the start of the simulation.

• A: No installation (no implementation of interaction policies)

• E: Near the central station, 0.5 km south and 0.5 km east of it

Table 2. Value of each indicator at the initial state and the start of the simulation.

	Representative transport method				CO ${ }_2$	Travel	SD
Scenario	Walk	Bicycle	Railway	Vehicle	emission	time	E-W	N-S
Initial State	0%	0%	100%	0%	17.2%	33.7 min	8.0	8.0
A	1.3%	2.4%	7.3%	89.0%	100%	9.3 min	22.9	9.8

Four interaction promotion coefficients were assumed: 0, 10, 20, and 30. These experimental Scenarios are denoted as ‘Scenario SA’ and ‘Scenarios SE0–SE30’ by combining ‘S’, the initial letter in ‘sprawl’, and the interaction promotion coefficient.

4.2. Results of experiment 1

Table 3 presents the results of the quantitative indicators of Scenarios SA and SE0–SE30. Figure 5 depicts the final distribution of residences in those Scenarios. In Scenario SA, the number of households using privately owned vehicles increased to 95% and dispersed habitation around the CBD expanded. In Scenarios SE0–SE30, although interaction policies were implemented, the households using privately owned vehicles did not shift to other means of transport and there was no improvement in dispersed habitation. Overall, in all the Scenarios, as in Scenario SA, the number of households using privately owned vehicles increased and dispersed habitation intensified.

Figure 5. Final distribution of residences in Experiment 1.

Table 3. Results of each indicator in experiment 1.

	Representative transport method				CO ${ }_2$	Travel	SD
Scenario	Walk	Bicycle	Railway	Vehicle	emission	time	E-W	N-S
SA	1.0%	2.3%	2.4%	94.2%	90.5%	6.5 min	27.5	11.6
SE0	1.0%	1.5%	2.4%	95.1%	97.4%	20.4 min	22.2	14.3
SE10	1.2%	1.2%	2.2%	95.4%	97.2%	20.0 min	22.2	14.1
SE20	1.0%	1.4%	2.2%	95.4%	98.5%	20.0 min	22.5	14.2
SE30	1.1%	1.3%	2.5%	95.2%	99.0%	20.2 min	22.7	14.1

5. Experiment 2: simulation assuming the introduction of tram

5.1. Conditions for experiment 2

This section reports the simulation experiment that assumed the introduction of tram. Following the ‘Karlsruhe model’ (De Bruijn and Veeneman 2009), it was assumed that the tramline would use the existing railway tracks. Specifically, as Figure 6 indicates, three tramlines were radially installed from the central station and their ends passed through some of the dispersed residences that surrounded the CBD. As is done for conventional tramlines, stops were installed at 400 m intervals on each route.

Figure 6. Schematic of tramlines.

The household agents could choose a linked trip with both the railway and tram or a linked trip combining the tram with walking or cycling (as shown in Figure 7). As in case of the previous experiment, this one was conducted under conditions in which two location options for the HUB: A and E, were combined with four interaction promotion coefficients. These experimental Scenarios are denoted as ‘Scenario SAt’ and ‘Scenarios SEt0–SEt30’ by combining ‘S’, the initial letter in ‘sprawl’, and ‘t’, the initial letter in ‘tram’ and the interaction promotion coefficient. Scenario SE30t+, which doubled the execution period of Scenario SE30t, was also implemented.

Figure 7. Linked trips using tram.

5.2. Results of experiment 2

Table 4 presents the results of the quantitative indicators for Scenarios SAt and SEt0–SEt30. Figure 8 illustrates the final distribution of residences in those Scenarios. In Scenario SAt, the number of households using privately owned vehicles increased and dispersed habitation around the CBD expanded, similar to Scenario SA. In the Scenario SEt series, from Scenario SEt0 to SEt20, the number of households using privately owned vehicles increased and dispersed habitation expanded, similar to Scenario SAt. However, in Scenario SE30t, in which interaction policies were encouraged, more than half of the households that used privately owned vehicles shifted to tram use. The total number of households using the tram reached approximately 50%. Owing to this shift, total CO ${ }_2$ emissions were reduced by more than 40%.

Figure 8. Final distribution of residences in experiment 2.

Table 4. Results of each indicator in experiment 2.

	Representative transport method						CO ${ }_2$	Travel	SD
Scenario	Walk	Bike	Rail	Vehicle	Rail+Tram	Tram	emission	time	E-W	N-S
SA	0.8%	1.3%	1.5%	93.4%	1.3%	1.7%	89.3%	7.5 min	27.3	11.7
SEt0	0.7%	0.9%	1.7%	94.1%	1.5%	1.2%	95.4%	20.9 min	21.6	14.2
SEt10	0.6%	0.6%	1.7%	94.5%	1.4%	1.2%	97.1%	21.1 min	21.9	14.2
SEt20	0.8%	0.7%	1.4%	91.8%	1.9%	3.3%	94.8%	21.3 min	21.9	14.3
SEt30	2.3%	0.9%	4.1%	44.9%	12.3%	35.6%	54.2%	37.4 min	22.5	14.2
SEt30+	0.8%	0.9%	1.4%	10.0%	19.4%	67.5%	20.6%	43.7 min	20.5	12.9

Furthermore, in Scenario SEt30+, in which the execution of Scenario SEt30 was extended, the number of households using the tram reached approximately 90% and total CO ${ }_2$ emissions significantly decreased, by approximately 80%.

Regarding the distribution of residences, two clusters emerged. One was centred around a suburban station comprising households that used a combination of railways and trams for commuting (approximately 20% of all the households). The other was a cluster along the tramlines, stretching from the centre of the CBD to the periphery, comprising households that used only the tram for commuting (approximately 70% of all the households).

6. Discussion

6.1. Discussion of experiment results

The combination of the appropriate placement of a hub facility that urban residents can conveniently access (near the central station) and policies to promote human interaction was demonstrated to be exceptionally effective in restraining the use of privately owned vehicles and dispersed habitation when work and life were separated at the time of policy implementation (Nagai and Kurahashi 2021). However, when dispersed habitation was at an advanced stage, these policies could not restrain its increase or the rise in the use of privately owned vehicles. It suggests that the worsening of the urban environment due to low-density suburban habitation and lifestyles, dependent on privately owned vehicles, is deep-rooted and difficult to reverse.

Additionally, this policy measure, when used alone, achieves the results described above when the tram service connects the CBD to a group of residences distributed in low-density surrounding it. By contrast, when the combination of tram use and appropriate placement of a hub facility was augmented by interaction promotion policies in the area in question, we observed a rapid shift from privately owned vehicles to trams as if a ‘phase transition’ had occurred. Moreover, CO ${ }_2$ emissions were reduced and dispersed habitation improved. If these policies are maintained over long periods, a significant improvement in the urban environment can be achieved. Thus, to return to the first of the questions we had posed in the Introduction, improvements in dispersed habitation in peripheral urban areas could be modelled based on the micro–macro linkage between spatially explicitly autonomous active agents and a planar urban scheme. This fact also suggests that a combination of the appropriate placement of the hub facility, interaction promotion policies, and the introduction of tram can ensure the effective management of sustainable urban growth.

Nevertheless, these policies will remain ineffective unless they are implemented on a relatively large scale. Interaction promotion policies induce walking in urban areas through incentives such as convenience, safety, and a rich experience. Similarly, many successful instances of tram introduction combine the tram with other policies that are beneficial for pedestrians. For example, in Karlsruhe, Germany, parts of tramlines were turned into a transit mall, which restricted car traffic (De Bruijn and Veeneman 2009). In Toyama, Japan, efforts were made to revitalise the shopping area along the tramline, which is far from the central station (Oba et al. 2008).

Thus, our experiment demonstrated that combining the introduction of the tram with policies that, for instance, induce pedestrian activities before and after tram use, significantly improves the urban environment. Here, regarding our second question in the Introduction, we can answer that residents’ short-distance public transport use and walking in the city centre improved dispersed habitation. Furthermore, for the third question, interaction promotion policies, which are collaborative efforts for the attraction of activities and closely integrated and cooperated with transportation, induced such behaviour of residents. As the implementation of interaction promotion policies requires collaborative efforts and costs from the local area, it is not simple, given the consensus building required. However, if we consider that not only urban residents but also businesses in the area would benefit from the liveliness induced by the accumulation of pedestrians, then the aim is not entirely unrealistic.

Regarding the distribution of residences in the same Scenario, the cluster of residences that emerge along the tramlines diverges from the original polycentric and rational land use based on the separation of work and life. However, it can be positively evaluated because of the resulting compact urbanisation.

• First, if a form of ‘living in the city’ is realised in which short-distance public transport plays a major role and people live close to their workplace, residents will be liberated from the problems posed by rush hour and air pollution.

• Second, mixed land use results in a wide range of social activities in the city centre, which supports its revitalisation. Therefore, the concentration of different activities will lead to the ‘urbanisation economies’ – a complementary economic effect (Krugman 1996; Beeson 2017) – and a creative urban environment (Knudsen et al. 2008; Malik et al. 2015; Belitski and Desai 2016).

Regarding the application of the TOD model at the regional city scale, our experiment revealed both a limitation and a vindication of the basic principles and policies of TOD. To combat an existing area of dispersed development and car dependency, a coordinated package of TOD as a site-specific development strategy and collaborative measures needs to be ‘retrofitted’. The first of the minimum necessary elements for the success of the ‘retrofit’ was the introduction of trams and high-quality public transportation. The second was what we called interaction promotion policies, which refer to the attraction of activities to areas around transportation nodes, that is, land use closely integrated and cooperated with transportation.

6.2. Testing the validity of the simulation model

When simulating real systems that are considered complex, including cities, abstracting the model as much as possible enables capturing the essential factors (Axelrod 1997; Epstein 2006). Therefore, the validity of the model could be ameliorated not only by including a few of the new parameters or changing their initial values but also by excluding a few of them. Additionally, the validity of the ABM needed to be tested to investigate whether it captured the basic characteristics of the real system, owing to the nature of emergence in a complex, self-organising system (Wu 2002; Drchal et al. 2016). Pattern-oriented modelling is an effective method to test validity in this case. It evaluates a model’s validity by examining whether the emergent pattern output, which is not directly defined as a parameter setting, recreates the observational patterns that characterise the real system, which is the subject of modelling (Railsback and Grimm 2011).

In Figure 9, the shift in the urban area in Japan, over 40 years (1970–2010) (yellow line) is superimposed on the shift in the residential area in the model over the 20 relocation cycles (equivalent to 40 years) from the initial state discussed in ‘Section 4. Experiment 1’ (blue line). The two show extremely similar shifts, indicating that the result of the Scenario can be regarded as a demonstration that urban areas have consistently expanded during Japan’s rapid economic growth (Eaton and Eckstein 1997). Additionally, the distribution of residences in the Scenario considerably changed from a zoning schema, which separated work and life, into dispersed habitation, in which most residences surrounded the CBD, as shown in Figure 4. This situation can be regarded as an example of the growth process of concentric, low-density, suburban habitation in accordance with the single-centred urban model that Alonso (1964) introduced and which has been supported by many subsequent studies.

Figure 9. Shift in urban area in Japan and residential area in the model.

Furthermore, most of the household agents living in a dispersed manner changed their means of transport from railway to privately owned vehicles. It can be regarded as a reflection of the tendency in regional Japanese cities during the same period when the major means of transport shifted from railways to privately owned vehicles, saturating the road networks (Ministry of Land, Infrastructure and Transport 2017). Additionally, Scenario SAt in ‘Section 5. Experiment 2’ demonstrates that introducing the tram to connect the CBD to a group of dispersed residences surrounding it could not, by itself, restrain the increase in dispersed habitation and privately owned vehicle use. It may explain why the development of public transport in suburban, low-density residential areas did not prevent sprawl or the use of privately owned vehicles in large cities such as in Atlanta, US (Bertaud and Richardson 2004) and Madrid, Spain (Garca-Palomares 2010).

This model does not aim to precisely recreate a real city but analyses the highly abstract mechanism of urban dynamics through a small number of simple elements and rules. Representatively, possible choices for the agents in relocation within the system were constrained mainly by urban planning laws or policies constraints on the basis of interaction promotion policies and introduction of tram. The findings of the modelling are based on such large degrees of freedom available to the agents. Nevertheless, several social phenomena (observational patterns) that were not directly built into the model were recreated. It suggests that the model, its setting, and its experimental results are sufficiently valid to examine the behaviour of a real city.

7. Summary

7.1. Research outcomes

This study proposed an exploratory urban morphology ABM to simulate a combination of the establishment of a hub facility that urban residents can conveniently access, policies to promote human interaction around it, and the introduction of tram that impacts dispersed habitations in peripheral urban areas. The stylised abstract urban model, which includes autonomous agents, provides insight into the micro–macro linkage in dispersed habitation, which is not desirable in any urban system.

A significant contribution of the proposed model is its introduction of a new perspective on how to improve the urban environment, by inducing a shift from dispersed habitation to concentrated habitation in a bottom-up manner through behaviour at the micro level, that is, via productive human interactions in an urban setting. Specifically, the model experimentally demonstrated a trade-off between creation of high-density active travel by tram and on foot within an existing area of dispersed development and increased dispersed habitation. It implies a vindication of principles and policies of TOD, which underlines the integration and cooperation of transportation and land use.

7.2. Future challenges

One of the ways of developing the model is to introduce demographics. In this study, the number of households was kept constant throughout the model execution period; however, in reality, demographics differ from one city to another. For example, in Europe, urban growth has been discussed in conjunction with population increase. By contrast, in regional Japanese cities, depopulation, as a consequence of low birth rates, and the ageing population have become major issues. The influence that such demographics would exert on policies for restructuring urban spaces is yet unknown and deserves further exploration.

Additionally, there may be concerns over the assessment whether the model settings and results can be applied in real-world situations – a problem frequently encountered when simulation is applied in the social science field. For example, property law might significantly affect the results if the heterogeneity of households in income and the real-estate market, which were not implemented for simplicity in this model, were considered. The same might be said for land availability, which, in the real world, varies according to geographical conditions and previous development. At future stages of the model development, these potential issues would need to be addressed.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the JSPS KAKENHI [21H01561,22K14445];The Obayashi Foundation;Japan Institute of Country-ology and Engineering Grant;

Notes on contributors

Hideyuki Nagai

Hideyuki Nagai received his Ph.D. degree from University of Tsukuba, Japan, in 2018. In 2021, he joined Kyoto Arts and Crafts University, where he is currently a fixed-term lecturer. His research interests include urban and architectural design and computational social science.

Setsuya Kurahashi

Setsuya Kurahashi received his Ph.D. degree from University of Tsukuba, Japan, in 2001. In 2006, he joined University of Tsukuba, where he is currently a professor of Graduate School of Business Sciences. His research interests include social simulation, artificial intelligence and history.

Notes

1. Calculated by multiplying the CO₂ emissions per passenger per transport distance for privately owned vehicles and the railway (Ministry of Land, Infrastructure and Transport 2021) by the total travel distance of each transport method of all the household agents.