2022 Volume 10 Issue 4 Pages 55-78
There has been much discussion about the shift to compact cities in the last decade. Transit-oriented development (TOD) has been introduced in many countries to promote compact cities. However, low density occurs with heterogeneous and random spatial patterns. It impedes the creation of people-friendly, sustainable cities, making it challenging to attract the investments necessary to create a compact city’s core. This phenomenon is a serious issue during periods of population decline. Therefore, this paper clarifies the occurrence pattern of low-density districts around railway stations and their related spatial characteristics to understand better how to reorganize transit-oriented urban structures in a society with a declining population. We have observed the population demographic and household demographic change beside railway lines of the Keihanshin conurbation within the Kyoto-Osaka-Kobe metropolitan region, the second-largest urban area in Japan. Our research questions are “Do low-density populations exist in railway station areas?”, and “If yes, are they concentrated at a particular point or dispersed?” The results of our research are as follows; 1. Some railway stations are experiencing low density growth in their surrounding areas. 2. Thirteen patterns of low-density appearance in the station sphere are observed. 3. Stations with low densification disperse, but at a relatively lower ratio in the inter-urban area. 4. There are three primary routes for low density progression. These results will be helpful in future urban planning, including site optimization, and can help attract facilities and mutual support between regions.
In the last decade, a shift to compact cities has been frequently discussed, starting with the use of sustainable urban structures to counter environmental problems (Kaido, 2001). The adjustment and redesign of expanding urban residential areas have become necessary in most developed countries over recent years (Hamel and Keil, 2015; Kötter, 2019; Reicher and Hesse, 2015), requiring optimising scale, facilities, and infrastructure to support current population structures (Dunham-Jones and Williamson, 2011). These are important issues, especially when population growth shifts from stagnation to decline. The compact city policy is based on two contexts of sustainable development: reducing environmental impact and optimizing the built environment based on the size of the population (OECD, 2012). Thus, most governments and municipalities in developed countries are attempting to realize these changes (ibid). In Germany, the Stadtumbau Ost program has been enacted to systematically reduce the size of cities with these objectives in mind (Hattori, 2016).
Transit-oriented development (TOD) has been introduced in many countries in conjunction with compact city policies (Carlton, 2009). The purpose of TOD’s is to create compact and walkable urban spaces centred on public transportation stations and using public transportation networks to connect regions, which fosters a healthy, low-carbon living environment (ITDP, 2017). Moreover, a mixture of land uses is often regarded as equally important, to reduce traffic in the first place (e.g., Bertolini (1999)). This change is critical in countries with advanced car societies (ESCAP and UN Habitat, 2019). For example, Japan has been developing public transportation since its urban expansion stage in early 1900s (Kadono, 2000; Uma, 2016), but has been motorizing since the second half of the 20th century (Miura, 2004). Currently, the Japanese government and its municipalities are reconsidering the nature of urban scale and public transportation, as well as, more specifically, the scale of cities and the nature of public transportation in a society with a declining population (Suzuki, 2015a, 2015b).
Under these circumstances, Japan formulated a Location Normalization Plan in 2014. The concept of the “Compact City and Network” is to shift the city to an urban structure suitable for future population size, considering sustainability (MLIT, 2015). In such a plan, residence-induced zones are established to incentivize people to relocate (Yoon, 2020), and railway stations and railway lines tend to be used as guidance points for these zones (Kadono and Matsune, 2021).
However, low density areas in the outer edges of conurbations, suburbs, and city centres are not evenly spaced and have heterogeneous and random occurrence (Aiba, 2014). This is called an urban perforation or hollowing-out effect and is a form of localized density reduction (Hollander, Pallagst et al., 2009; Rink and Siemund, 2016). This urban perforation impedes the creation of people-friendly, sustainable cities (Denis, Cysek-Pawlak et al., 2021). Moreover, areas with induced and concentrated functions create low use and unused spaces. The value of the area then declines, making it challenging to attract necessary investment (Japanese Ministry of Land, Infrastructure, and Transport (MLIT) (MLIT City Planning Division, 2018)). In addition, it is stated that urban perforation impedes compact city policies (ibid).
This type of low density can also occur around railway stations, an outcome for future TOD-based compaction of cities. Even in the vicinity of highly centralized stations, such as those where limited express trains stop or multiple train lines intersect, individuals from younger generations may not tend to reside (Aoki and Kadono, 2020b). In addition, if residential areas are along the same railway line, they often belong to different phases of household demographic transition. Some are under ongoing population outflow trends (Aoki and Kadono, 2020a).
In the population decline stage, it is difficult to develop new TODs on a large scale. Therefore, it is necessary to consolidate existing residential areas to be based around existing stations and gradually convert them into a compact urban structure. However, at the free will of people and markets, low-density areas can occur randomly. Railway stations predetermine residence-induced zones the reorganizing of urban structure, yet the random occurrence of low-density districts, defined as urban perforation, is a severe issue for reorganizing and optimizing urban scales centred on railway stations. Thus, it is important to assess low-density districts around railway stations to identify the possibility of transit-oriented reorganization.
In the field of TOD, in addition to the use density around stations, the diversity of land use is also important. However, in a declining population, it is difficult to redevelop every railway stations' sphere due to constraints, such as financial ones. In fact, with the exception of a few terminal stations, they have not been designated as urban function guidance areas in Japan’s Location Normalization Plan. For this reason, this paper focuses on the consolidation of residential areas, which is the first and most important aspect of urban restructuring based on public transportation in Japan
Japanese conurbation expanded in connection with public transportation, especially railway lines. However, Japan’s national population size has been decreasing since 2010 (National Institute of Population and Social Security Research (IPSS) (IPSS, 2019)). Thus, most municipalities are confronting the urgent tasks of reducing bloated scale of residential areas and compacting them to centre around railway lines and stations. The Keihanshin conurbation is a group of municipalities with strongly connected urban centres within the Kyoto-Osaka-Kobe metropolitan area, which is the second largest conurbation in Japan. However, even such an area is experiencing population decline.
This paper assesses the recent situation of the Keihanshin conurbation and clarifies the occurrence pattern of low-density districts around railway stations and their related spatial characteristics in an era of depopulation, to consider which railway lines and stations have the possibility of being set as urban cores for future reorganization. We posed the following research questions:
1. Do low-density populations exist in railway station areas?
2. If yes, are they concentrated at a particular point or dispersed?
Related worksImplementing TOD, including compact cities, can significantly promote sustainability. However, conurbations must manage urban perforations while dealing with depopulation. In that context, many researchers have discussed effective policy structures and methodologies to create urban form compactness (Bibri, 2020; Bibri, Krogstie, et al., 2020; Korthals Altes and Tambach, 2008), as well as the compact model’s contribution to sustainability (Iizuka, Xuan et al., 2020; Wang, Liu, et al., 2021) and compact city theories’ effect on urban expansion in areas with population growth (Xu, Zhou et al., 2020). In addition, studies clarifying the possibility of proactive land use within a station’s immediate sphere and accessibility to a station in line with walkability have been conducted (Lyu, Bertolini, et al., 2020; Wenner and Thierstein, 2021). These are critical perspectives when questioning the pros and cons of compact city policies and TOD theory.
On the aspect of evaluating ongoing TOD, Staricco and Vitale Brovarone (2018) examined the role of regional planning for TOD in two European cities and found a level of hardly accomplishable deep coordination of land use and transport. Additionally, TOD can be capitalized into land values, however, the amount and its impact have a wide range that differs by TOD plan (Higgins and Kanaroglou, 2016). On the other hand, a study on the Asian TOD context gives a hint of success factors, such as a shift from highway-based zoning to transit-oriented zoning and the creation of institutional mechanism for the multi-stakeholder in station area development (Kidokoro, 2020). Moreover, a study on typifying stations beside their surrounding land use status has been attempted in various cities (e.g., Higgins and Kanaroglou, (2016), Song and Deguchi, (2013)). Furthermore, some studies have found that population growth areas form outside of station spheres, contrary to the TOD trend (Ono and Sato, 2018) and the characteristic of residents having a higher demand for living beside stations (Pongprasert, 2020).
Recent research on low-density landscapes has attempted to understand patterns of depopulation and the drivers of depopulation. It is particularly important to use spatial indicators to explore and simulate urban growth and degradation patterns), but there are few spatial indicators to study urban shrinkage and the occurrence of low-density areas, and there is a need to develop new or mixed indicators (Reis, Silva et al., 2015). Moreover, once vacant land accumulates, urban perforation tends to continue (Usui and Perez, 2020). Several major pathways lead to vacant land, including unplanned political and economic development, inadequate and low-quality infrastructure, and an aging population (Jeon and Kim, 2019). Furthermore, once low-density among the younger generation occurs, it can affect adjacent areas and lead to additional low density (Aoki, 2022).
In addition, studies on the demographic changes in the station sphere have primarily focused on the elderly population and population growth. For example, Ito, Nakagawa et al. (2011) showed that while populations reach higher density around stations with more frequent rail services, the percentage of the elderly population is lower. Moreover, even with the same public transportation system, population growth and retention are expected to be higher on rail lines than on bus lines (Ochiai, Makimura, et al., 2019).
Some studies have analysed the reasons for migration to station spheres and found that housing prices and the surrounding environment are more likely to attract attention than access to transportation (Lund, 2006). However, these analyses mainly focus on urban centres and areas where population growth continues on an urban scale. Some studies have examined areas with declining populations, but they have focused on third-sector rail lines in regional cities, looking at the sustainability of rail lines rather than the region (Asami et al., 2019). Population growth is not solely a problem for regional cities; it can also occur on a metropolitan scale (D. Haase, A. Haase et al., 2014).
Thus, the promotion of TOD and compactness has been proven to have a certain value and potential. In addition, the transformation of stations and their surroundings due to development has been analysed and policy assessments have been conducted. However, they fail to take into account the possibility that low-density areas may emerge around stations in conurbations undergoing population decline.
Analysis of the location and causes of low-density development is also underway, but there is no research that focuses on low-density development in the station sphere itself. As mentioned above, population decline and urban perforation are important issues for future compactness, which is also true for metropolitan areas. Observing demographic changes in the railway station sphere, especially low density trends, can be an essential guideline for future urban planning. Thus, this paper examines low densification occurring within the station sphere.
This study examines the demographic changes and current situation of the station sphere with a 1 km radius of each railway station. The distance of a 1 km radius is based on the identification of the TOD compactness indicator as stated by ITDP, UN-Habitat, and other institutions (ITDP, 2017). The range of walking distances in TOD has been discussed and examined to various extents (e.g., Ewing and Cervero (2010); Calthorpe Associates (1992)). In Japan, the most popular range of walking distance in public opinion polls is about 1000m, which is suitable for the range employed in Institute for Transportation and Development Policy (ITDP). Furthermore, a radius of 1 km was selected as the best and smallest range for data convenience, since ranges of 500 m or less cannot be aggregated. There are three sequential steps in the framework of this analysis.
The Keihanshin conurbation expanded significantly from the 1960s to the 1990s. Figure 1 shows that densely inhabited districts have sprawled into suburbia during this time, mainly along the railway lines, however, its population began to decline earlier than the Tokyo and Chukyo areas (Okada, 2007), beginning in the 1970s (IPSS, 2021). As such, population decline in this conurbation is likely to continue, making its railway station sphere suitable for observation of the occurrence of low-density populations.
Based on the spread of densely inhabited districts and the ratio of residents who commute to the city centre as of 2015, the Keihanshin conurbation is defined as the area with a ratio of 10% or more of the total number of employees in the three cities (Kyoto, Osaka, and Kobe) to the total number of employees in the aggregate data for each municipality in the 2015 national census. In this study, the central urban area was the city ward, positioned as the city centre in each urban master plan. Kobe’s urban master plan defines the city centre as part of the existing urban area south of the Rokko mountain range, therefore, wards that fall within the city centre with a population centre south of the Rokko mountain range were defined as Kobe’s centre. Suma ward was excluded from urban centres, although it has a built-up area south of Rokko Mountain. The Osaka city area was determined using the master plan of the Osaka prefectural city planning (Figure 1).
There exist 1109 railway stations in the conurbation. This paper observes all railways except high-speed rail (Shinkansen) since TOD conceptually focuses on regional community life in one aspect (Jamme et al., 2019). Then, we analysed the station sphere of railway stations with the 500-meter-mesh national census data to focus on the population dynamics of the station sphere. The mesh is considered part of the station sphere of the corresponding station only when the centre point of a 500m mesh falls within a 1 km radius of a railway station.
National census data for 2005 and 2015 were used for the analysis. No residential areas have been systematically developed in the Keihanshin conurbation since 2005, and most residential areas were developed by 2015 (MLIT, 2018). In addition, it took several years to recover from the Great Hanshin-Awaji Earthquake of 1995, and some households were still living in temporary housing until 2000 (Kobe Newspaper, 2000), so we assumed the possibility of influence from the Great Hanshin-Awaji Earthquake on the 2000 census. In addition, Japan's population began to decline in 2008. By using data from 2005 and 2015, we hoped to capture the changes from before an area’s population begins to decline, to after. As the primary mode of transportation, the railway usage rate was 20.8% as of 2010, in the middle of the year analysed (Keihanshin Metropolitan Area Transportation Planning Council, 2012). Compared to the 31.3% for car usage, it is relatively lower. However, the ratio of railway usage was 18.9% in 2000, indicating a slight increase over the past ten years. This survey also includes rural areas outside of the target region. The percentage of rail dependence in the Keihanshin conurbation, which has a more developed rail network, is likely higher than in this report.
*Densely Inhabited district (DID) is a district containing basic 1km squared mesh units with a population density of 4,000 or more and adjacent to each other, forming an area with a total population greater than 5,000. The Japanese Statistics Bureau created this definition to capture the urbanized area.
First, an overview analysis of the demographic changes in the sphere was conducted. All objective railway stations and their station spheres were estimated using QGIS. All 500m-mesh census data were connected to each railway station. Some of the mesh was overlapped by two or more stations, and these mesh data were attached to each station.
It has been confirmed that the number of household members decreases during the aging and population decline phases. Japan, in fact, is experiencing a shift to single-person households in both younger generations, due to the trend toward late marriages, and the elderly due to the passing away of a spouse. Given these factors, we believe that population increase/ decrease is not necessarily directly related to the flow of new households into and out of the country. Therefore, the change ratio in the population and number of households of all the 500m meshes of the Keihanshin conurbation were calculated and divided into quartiles. The ratio considered was from 2005 to 2015.
After determining into which group the 500m meshes located within all the station spheres should be sorted, we analysed what percentage of the meshes belonging to each quartile group were located within the station sphere. This was done to determine whether the number of people and households in the station sphere tended to be larger (or smaller) than in other areas.
Next, we observed the difference in the demographics of the station sphere depending on the typology of the railway station within the conurbation. The classifications of train stations in the context of TOD vary from region to region. For example, Denver has five types (Denver, 2014): Downtown, Urban Centre, General Urban, Urban, and Suburban, while Kedungsepur has four types (Ramadhan and Pigawati, 2019): Urban Core, Urban General, Suburban, and Rural. In this paper, however, the stations are classified according to Paul Romer's current perspective on urban science, i.e., that the city should be treated as a unit of analysis (Romer, 2013). In Japan, the former Japanese National Railways (now Japan Railway) and many private railways have railway networks, some of which run parallel. The development of residential areas has accompanied the expansion of these railway lines. These residential areas are bedroom communities in the city centre. Thus, it is not easy to observe their characteristics under a scenario of separation from the city centre where offices, commercial facilities, and leisure facilities are concentrated.
The Keihanshin conurbation consists of a multi-nuclear structure with three urban centres: Kyoto, Osaka, and Kobe. For this reason, this paper will classify stations in terms of 1) Urban: those located within the city centre, to begin with, 2) Inter-Urban: those with easy access to two or more city centres, 3) Suburb: those located on the outer edge but with relatively high accessibility to the city centre, and 4) Branch: those located within the urban area but with relatively low accessibility to the city centre.
The station classes are divided into the following (Figure 2):
To overview population concentration tendency around railway lines, we observed quartiles of both population-level and household-level demographic transition. In the case of both population- and household-level transition, we estimated the percentage of meshes that fell into each quartile group at the individual station class. Each class's ratios were compared by pairwise comparisons using the Wilcoxon rank-sum test and the Kruskal-Wallis rank-sum test.
Finally, we observed the current status of each 500m mesh through the correlation of the population demographic change ratio and the household demographic change ratio. In both ratios, the value 1.0 determined whether there was an increasing or decreasing trend in 2015 compared to 2005. Therefore, as shown in Figure 3, we defined each quadrant based on the combination of the population demographic change rate and the household demographic change rate. According to the quadrants, entire meshes that fell under each station class were calculated.
Second, we clarified the occurrence pattern of low densification in the station sphere. To examine the occurrence, the Local Moran's I statistical method, which is often used to identify local clusters and spatial outliers, was used ( Anselin, 1995). It determines the relationship between a mesh and its neighbouring meshes and consists of the component in the double sum that corresponds to each observation i with j as the neighbouring meshes:
For the Local Moran's I statistic, this study defines the low densification coefficient. The coefficient is composed of the combination of the demographic change ratio and the household change ratio. These two aspects are necessary to observe the low densification tendency in a particular mesh. The formula for the low-densification coefficient ( v) at the mesh ( i) is as follows:
As a result, all meshes are plotted on the horizontal axis in Figure 4 , which shows the standardized variable of a particular 500m mesh, while the vertical axis shows the standardized spatial lag variable. Neighbouring meshes contain the variable used in the spatial lag. The first quadrant in the figure shows the districts with high values for both the specific mesh and its adjacent areas. A mesh in the second quadrant has a low value, with high values in the adjacent meshes. The third quadrant has low values for both itself and its adjacent districts. The fourth quadrant represents a mesh with a high value itself and with low values in the adjacent meshes. We included only those 500m meshes for which statistical significance was demonstrated by the Local Indicator of Spatial Association (LISA).
The results obtained by Local Moran are plotted on a map, and the distribution within each station sphere is considered. Since this paper examines the low densification occurrence, we mainly focus on the low-low and low-high of the Local Moran analysis. The stations are then classified according to the flowchart shown in Figure 5 , based on the distribution of low-low, low-high, high-low, and high-high.
Observing the characteristics of station spheres that are "At Risk" and "Low-Densification"Finally, we conducted a statistical analysis to clarify the characteristics of railway stations that are at risk of or are currently experiencing low densification. This step defines “low-densification” stations as Pattern ID 1 to 5 in Figure 6 and the “at risk” stations as Pattern ID 1–7, 10–11, and 13–16. To consider these characteristics, this study uses decision tree analysis. The explained variable is the binary data of Yes/No for “At Risk” and “Low-Densification.” The explanatory variables are presented in Table 1.
Changes in household structure were used because life stages may encourage migration (Aoki and Kadono, 2020a). Since home ownership patterns have been changing recently (Hirayama, 2011), we focused on housing and ownership patterns. To consider economic aspects, public land prices for the residential properties nearest to each station were used. In addition, recent redevelopment projects outside stations have attracted residences and offices, so the change ratio of the number of employees in each business category was examined, where occupational classification is based on the Japan Standard Occupational Classification.
Finally, the low densification coefficient outside the station sphere was also used as an explanatory variable to examine the relationship with the increase in residents outside the station sphere.
| Ratio Type | Category | Variables |
|---|---|---|
| Ratio of Total | Demographic | Relative households (2005), Households consisting of only husband and wife (2005), Households consisting of husband and wife and children (2005), Households other than nuclear family (2005), Households with under 6 year olds (2005), Households with under 18 year olds (2005), Households with 65 year olds and over (2005), Single-family house (2005), |
| Type of House | Row house (2005), Apartment house (2005), Owner-occupied house (2005), Rented house (2005), Apartment house 1-2 stories (2005), Apartment house 3-5 stories (2005), Apartment house 6-10 floors (2005), Apartment house 11 stories or more (2005) | |
| Occupation | Agriculture and Forestry (2005), Fishing (2005), Mining and Quarrying of Stone (2005), Construction (2005), Manufacturing (2005), Electricity, Gas, Heat and Water Provision (2005), Information and Communication (2005), Transportation and Postal Services (2005), Wholesale Trade and Retail Trade (2005), Financial Industry and Insurance (2005), Real Estate Leasing for Goods and Services (2005), Lodging, Food and Beverage Services (2005), Medical Services and Welfare (2005), Education and Learning Support Services (2005), Complex Services (2005), Employed persons, including directors (2005), Self-employed, including domestic workers (2005), Family Employee (2005) | |
| Transition Ratio | Demographic | Under 10 (2005-2015), 10s (2005-2015), 20s (2005-2015), 30s (2005-2015), 40s (2005-2015), 50s (2005-2015), 60s (2005-2015), Over 70 (2005-2015), Relative households (2005), Households consisting of only husband and wife (2005), Households consisting of husband and wife and children (2005), Households other than nuclear family (2005), Households with under 6 year olds (2005), Households with under 18 year olds (2005), Households with 65 year olds and over (2005), Single-family house (2005) |
| Type of House | Row house (2005), Apartment house (2005), Owner-occupied house (2005), Rented house (2005), Apartment house 1-2 stories (2005), Apartment house 3-5 stories (2005), Apartment house 6-10 floors (2005), Apartment house 11 stories or more (2005) | |
| Occupation | Agriculture and Forestry (2005), Fishing (2005), Mining and Quarrying of Stone (2005), Construction (2005), Manufacturing (2005), Electricity, Gas, Heat and Water Provision (2005), Information and Communication (2005), Transportation and Postal Services (2005), Wholesale Trade and Retail Trade (2005), Financial Industry and Insurance (2005), Real Estate Leasing for Goods and Services (2005), Lodging, Food and Beverage Services (2005), Medical Services and Welfare (2005), Education and Learning Support Services (2005), Complex Services (2005), Employed persons, including directors (2005), Self-employed, including domestic workers (2005), Family Employee (2005) | |
| Cohort Ratio | Demographic | Under 10 (2005-2015), 10s (2005-2015), 20s (2005-2015), 30s (2005-2015), 40s (2005-2015), 50s (2005-2015), 60s (2005-2015), Over 70 (2005-2015), |
First, we observed the demographic and household transitions in the entire objective conurbation and the station spheres. Figure 6 shows the distribution of quartiles for both the demographic and household ratios in the conurbation. The blank area on the map represents meshes which do not contain residences. The first quartile has the darkest colour, and it progressively lightens until it reaches the fourth quartile. Many meshes in peripheral districts fall under the first or second quartile; however, some meshes have lower ratios even near the city centre. Compared to the demographic, the household change ratio is more likely to have a lower ratio around city centres.
Figure 7 shows the percentage of each quartile located in the railway station sphere. In both cases, more than half of the meshes in the third quartile were located in the railway station sphere. On the other hand, only 20.0%–25.0% of the meshes in the first quartile were within the sphere. There was no significant difference between the second and fourth quartiles, which remained at approximately 45.0%.
We calculated the ratio of the quartile groups occupied by the relevant mesh for each station class, shown in Table 2. Regardless of the population or household demographics, “urban” has a higher ratio in the fourth quartile, while “inter-urban” has the highest ratio in the third quartile, and the other two classes have a relatively higher ratio in the first and the second quartiles. According to the Kruskal-Wallis rank-sum test, each change ratio significantly differed between the classes (Figure 8). Moreover, the urban and other classes had a significant gap in both change ratios. The inter-urban class had significant differences between “suburb” and “branch”, but only in the population-level demographic.
Finally, we observed the station classes’ current situation through population and household demographic change ratios. Figure 9 shows the resulting scatter plot; the percentages of meshes falling under each quadrant are shown in Figure 5. The urban and inter-urban classes have lower percentages of the density reduction mesh, and the plot distribution was generally linear, while in the other two, it was diffuse. More than half of the mesh in the urban class falls under the first quadrant, showing the demographic densification. Other classes had a relatively higher ratio in the second and third quadrants. In addition, about one-third of the meshes in the suburb and branch classes fall under the third quadrant, representing density reduction.
| Urban | Inter Urban | Suburb | Branch | |
|---|---|---|---|---|
| n=560 | n=800 | n=3038 | n=1905 | |
| Population-level | ||||
| First Quartile | 9.3 | 5.0 | 11.5 | 13.8 |
| Second Quartile | 17.3 | 24.9 | 29.1 | 28.8 |
| Third Quartile | 31.4 | 43.5 | 33.5 | 32.2 |
| Fourth Quartile | 42.0 | 26.6 | 25.8 | 25.2 |
| Household-level | ||||
| First Quartile | 13.2 | 9.8 | 15.1 | 16.6 |
| Second Quartile | 17.0 | 28.4 | 28.0 | 27.8 |
| Third Quartile | 30.5 | 40.4 | 31.8 | 32.0 |
| Fourth Quartile | 39.3 | 21.5 | 25.2 | 23.6 |
In the conurbation, we used Local Moran's I statistics to clarify the distribution of high-high (HH), high-low (HL), low-high (LH), and low-low (LL). The clusters of these meshes are shown in Figure 10. HH mainly existed along railway lines, not only within the city centre. LL and LH occur throughout the entire conurbation, including the city centre and edge. HL, on the other hand, tends to appear unevenly on the edge of the conurbation.
The percentage of each cluster that is located within the station sphere is shown in Table 3. We then estimated the number and percentage of the stations in each station class that contained LL or LH mesh within its station sphere (Table 4). Most meshes around station spheres fall under High-High or Low-High, indicating that reorganizing transit-oriented compactness is partially progressing. However, since more than a tenth of meshes are Low-Low, the population outflow trend in a cluster of meshes exists even within the station sphere. According to the percentage of LL and LH in each station class, it is inferred that those low-density occurrences are highly observable in the urban, suburban, and branch classes.
| Within the Station Sphere | |||
|---|---|---|---|
| Total | n | % | |
| High-High | 228 | 96 | 42.1 |
| Low-Low | 433 | 64 | 14.8 |
| Low-High | 406 | 132 | 32.5 |
| High-Low | 112 | 10 | 8.9 |
| Total | Urban | Inter Urban | Suburb | Branch | ||
|---|---|---|---|---|---|---|
| Number of Stations | 1234 | 251 | 128 | 596 | 259 | |
| Stations with LL and/or LH | ||||||
| Number | 234 | 73 | 12 | 94 | 55 | |
| Ratio | 19.0% | 29.1% | 9.4% | 15.8% | 21.2% | |
|
Average Number of 500m-mesh in each station sphere |
11.9 | 12.0 | 12.3 | 11.9 | 11.8 | |
| Ratio of meshes with LL and LH (%) | ||||||
| Max | 53.8% | 53.8% | 15.4% | 41.7% | 50.0% | |
| Average | 14.9% | 14.5% | 11.4% | 14.8% | 17.0% | |
| Min | 7.1% | 7.7% | 7.7% | 7.1% | 7.7% | |
| Stations with LL and/or LH mesh ratio | ||||||
| Less than 10.0 | 112 | 35 | 7 | 47 | 23 | |
| 10.0 to 30.0 | 112 | 36 | 5 | 43 | 28 | |
| More than 30.0 | 10 | 2 | 0 | 4 | 4 | |
Three new railway lines have been constructed from 2005 to 2015. However, these clusters appear only at one of the lines' nodes and in the central urban area station (Figure 11). This indicates that new railway construction in the urban area does not involve demographic growth or shrinkage.
Stations that contain LL or LH are classified according to the flowchart shown in Figure 6. Nine types of station spheres were extracted, as shown in Figure 12. Station spheres that do not contain any HH or HL are defined as “Low-Densification.” Even if a mesh with a tendency toward station sphere densification was included, it was judged to be “at risk” if it was in the furthest extension.
| Urban | Inter-Urban | Suburb | Branch | TOTAL | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| n=251 | n=128 | n=596 | n=259 | n=1234 | |||||||
| Whole Low | 1 | 2 | 0.8% | - | - | 2 | 0.3% | - | - | 4 | 0.3% |
| Half Low | 2 | 3 | 1.2% | - | - | 8 | 1.3% | 6 | 2.3% | 17 | 1.4% |
| Inner Low | 3 | 2 | 0.8% | - | - | 5 | 0.8% | 1 | 0.4% | 8 | 0.6% |
| Outer Low | 4 | 6 | 2.4% | 3 | 2.3% | 12 | 2.0% | 11 | 4.2% | 32 | 2.6% |
| Topically Low | 5 | 5 | 2.0% | 6 | 4.7% | 39 | 6.5% | 9 | 3.5% | 59 | 4.8% |
| Low with Outer High | 7 | 3 | 1.2% | - | - | 2 | 0.3% | 3 | 1.2% | 8 | 0.6% |
| Inner Low with Outer High | 13 | 17 | 6.8% | 2 | 1.6% | 2 | 0.3% | - | - | 21 | 1.7% |
| Outer Coexist | 14 | 15 | 6.0% | - | - | 3 | 0.5% | 5 | 1.9% | 23 | 1.9% |
| Local Low with Outer High | 15 | - | - | - | - | 8 | 1.3% | 9 | 3.5% | 17 | 1.4% |
| High with Inner Low | 8 | 6 | 2.4% | - | - | - | - | 2 | 0.8% | 8 | 0.6% |
| High with Outer Low | 9 | 6 | 2.4% | - | - | 1 | 0.2% | 3 | 1.2% | 10 | 0.8% |
| Inner High and Outer Low | 11 | 1 | 0.4% | - | - | 3 | 0.5% | - | - | 4 | 0.3% |
| Randomly Coexist | 18 | 7 | 2.8% | 1 | 0.8% | 9 | 1.5% | 6 | 2.3% | 23 | 1.9% |
| "Low-Densification" category | 18 | 7.2% | 9 | 7.0% | 66 | 11.1% | 27 | 10.4% | 120 | 9.7% | |
| "At Risk" category | 53 | 21.1% | 11 | 8.6% | 81 | 13.6% | 44 | 17.0% | 189 | 15.3% | |
| Total | 73 | 29.1% | 12 | 9.4% | 94 | 15.8% | 55 | 21.2% | 234 | 19.0% | |
Table 5 shows the number and ratio of patterns that fell under each station class. Except for the inter-urban class, the classes were not biased toward any particular pattern, and various low density development patterns were identified. In addition, the percentage of stations that fall under the “at risk” category is much lower in the inter-urban class compared to the other classes. The distribution of stations that fell into the “at risk” and “low densification” categories are shown in Figure 13. The stations in the “low densification” category are generally located in the inner part of the target metropolitan area. If the “at risk” category is included, stations in the peripheral area tend to be included. However, these categories have fewer stations on the lines running from central Kobe to the west and from central Osaka to the southwest along the coastline than along the lines extending in other directions. These two lines lead to Himeji City (west of Kobe) and Wakayama City (south of Osaka), respectively, which are core cities outside the metropolitan areas.
Classification tree analysis of low-density station sphereIn decision tree analysis, it is necessary to prevent the resulting tree from becoming more complex than necessary. Thus, we first observed the amount of change in the complexity parameter (cp) for each tree. In both analyses, the misclassification rate tended to level off after the cp reached 0.04. Therefore, we set cp = 0.04 as the threshold for both analyses. As a result, the five ndoes were analysed for “At Risk” and three for “Low Densification.”
Figure 14 shows the tree representing the result of “At Risk”. According to the analysis, [Ratio of Row House, 2005] is represented as the first branch point. The ratio of row houses in Japan is initially low, but in this relatively low case, [Ratio of total Households with. under 6year olds] is essential, and areas with many households with young children are more likely to be at risk. In areas with relatively large numbers of tenement buildings, the risk of low density tends to be higher when the increase in the percentage of employees involved in medical and welfare services is 189.791% or more, and the change in the amount of multi-family housing is generally less than 200%.
Two divergent points were identified in the analysis targeting low densification, where low density had already occurred (Figure 15). The first is [Transition. Ratio. 10s. 2005. 2015], and the second is [Ratio. of. total. Lodging. Food. and. Beverage. Services. 2005]. As the population continues to decline, there is a tendency for lower densities when the amount of change in teenage residents within the station sphere is under 68.475%. In addition, many areas with a large number of service-related workers as of 2005 have become low density.
The misclassification rate of the decision tree obtained in this study was 11.5% for “At Risk” and 8.4% for “Low Densification.” Although there is room for improvement in the explanatory variables, we believe that the results are generally appropriate for estimating the low density of the station sphere.
Looking at occurrence trends of low density, both the population and the number of households occur in spatially random distributions. Compared to the overall trend of decreasing population, the specific decline in the number of households appears to occur more frequently in city centres. Against this backdrop, the demographic trends in the vicinity of railway stations tend to fall into the third and fourth quartiles, with relatively high density being maintained or increased. However, about one in four or five of the first quartile are also in the station sphere, indicating a trend toward lower density along other lines than the urban class.
In this context, the results of the Local Moran analysis show that localized (LH) and clustered (LL) low densification are also occurring in the station sphere. In some cases, this type of densification occurs consistently throughout the station sphere, while in other cases, it occurs randomly in the centre or at the edges of the sphere. It has been argued that the occurrence of urban perforation is irregular regardless of whether it occurs in the city centre, inner and outer suburbs, or edges of a conurbation (Aiba, 2014). This study confirms that such randomness in perforation can occur not only at the conurbation level but also at the scale of the railway station sphere. In addition, a pattern in which low density occurs around stations and high density at the edges is observed, which does not necessarily mean that a compact residential space has been formed by defining the station as the centre.
In addition, low density in the station sphere is not concentrated in any particular pattern, and when including the “at risk” category, nine patterns were observed. This tendency is not only observed in the suburb and branch classes but also in the urban and inter-urban. However, there is a relatively lower possibility for inter-urban stations to fall into the “at risk” or “low densification” categories. Moreover, even in the suburb class, the number of stations in this category is low along the lines connecting core cities outside the objective conurbation to the city centre. Therefore, in a mortal society with depopulation, it is beneficial to assume railway lines that either connect two city centres or connect the city centre and core cities outside of the conurbation, as the central axis in planning a regional TOD structure in the future. In comparison, although stations are located close to the city centre, if one end of the railway line does not reach any core cities, there is a relatively high probability of low densification.
We estimated three main factors that could cause the station sphere to have low densification based on the two decision tree analyses. They are: row house demolition and use; the outflow of households with small children; and the availability of transportation to commercial and service facilities
Row houses in Japan have been rebuilt due to aging and a lack of fire protection (Fujita, 2014). However, even if they were converted to non-residential uses, such as medical and welfare facilities, or rebuilt into housing complexes, there is a high risk of low density. Although the rebuilding of row houses is a problem for owners, the district must consider a plan that anticipates using the space after rebuilding, if possible.
Even in station spheres with few such locations, low density can occur if there are many households with children under the age of six. There is a tendency for households to relocate when their children enter elementary school (six years old or older) to find a suitable educational environment for their children (Cho et al., 2019). If the same kind of change is occurring in the station sphere, it can be inferred that the priority of accessibility to public transportation is not that high. Furthermore, to achieve compactness centred on railway stations in the future, it will be necessary to develop a living environment suitable for raising children.
Finally, the tendency to fall into the low densification category is stronger in station spheres with higher numbers of commercial and service-related workers. This phenomenon can be attributed to the recent redevelopment of railway stations, their development into a place where people can gather, play, and enjoy their lives, rather than just a transportation node (Itami, 2021). In addition, such redevelopment occurred in relatively major forms in central urban areas, spontaneously giving us the idea of higher percentages of the "at risk" and "low-densification" categories in the urban class. However, as shown in Figure 14, some of the low densification stations are located close to each other. If such an overemphasis on the service industry occurs with multiple stations within a small area, competition for customers will likely occur between stations. Therefore, while the lower density associated with the change of use is understandable in some respects, differentiating stations by comprehensively considering neighbouring stations on the same line and nearby stations on other lines is necessary.
This study explored low densification occurrence typology and its characteristics within the station sphere in a society experiencing population decline. Our results confirmed that some railway stations are experiencing low density growth in their surrounding areas. The locations of these railway stations are independent of whether they are located in the city centre or the suburbs. When considering the scale of a conurbation, railway lines that connect city centres or run between the central and the core city can be the mainstay of TOD.
We also found that there are three primary routes for the progression of low densification. Based on these findings, it can be inferred that differentiation between stations, the development of a living environment where households with children can continue to live, and the guidance of land use after reconstruction can contribute to the prevention and revitalization of low-density areas. These results will be helpful in future urban planning, including site optimization planning, when aiming to attract facilities and mutual support between regions.
However, this study has some limitations. First, there is a limit to the type of spatial statistical data that can be obtained. Since the analysis was based on publicly available statistical data, we have not yet considered the effects of factors other than the explanatory variables shown in Table 1. It is therefore necessary to improve the accuracy of future analyses by conducting interviews with households relocating out of the station sphere.
This study was conducted as a case study of the Keihanshin conurbation. Since the results obtained may differ depending on a region’s characteristics, it is necessary to improve the generalization of the results by conducting multiple surveys targeting different conurbations in the future. These analysis methods can be applied in Japan and other regions if spatial statistical data on a small area or a mesh scale of approximately 500 m are available. In this respect, it can be said that this study is highly versatile. This study focused on the overview of demographic change at the scale of the conurbation, allowing "a juxtaposition of dynamics with existing structure at stations" to also be a future research topic by limiting stations to observe.
As a by-product of this study, the demographic growth trend outside the station sphere was also identified. This trend essentially represents a development that is contrary to the aims of TOD. Thus, clarifying the characteristics of these areas should continue. In addition, the latest National Census data available for 500m mesh is from 2015, so may be out of date. Even though the data for all of Japan have been published in 2020. Unfortunately, data at the small district level used in the analysis, will not be available for three to four years. Thus, 2015 is the latest available set that includes 500m-mesh data. However, the occurrence of the 2020 COVID-19 pandemic may skew the effects of multiple social backgrounds if all current data were analysed in one study. Therefore, it is necessary to re-analyse the impact of the pandemic on the formation of compact cities centred on train stations by referring to data from a few years after the pandemic subsides, such as from 2015 to 2025.
The authors declare that they have no conflicts of interest regarding the publication of this paper.
We would like to thank Editage (www.editage.com) for English language editing.
