The wider barrier effects of public transport infrastructure: The case of level crossings in Melbourne

As an enabler of land use changes, transport infrastructure has been widely documented in the literature. However


Introduction
Transport infrastructure plays a vital role in shaping and structuring urban economic, environmental, and social activities by enabling access to destinations (Xueliang, 2013). However, it can also pose a significant barrier for communities. For example, a motorway bisecting a community in two parts leads to a loss of community accessibility and cohesion. The literature refers to these unintended consequences of transport infrastructure as the barrier effects or community severance (Anciaes et al., 2016;Taylor and Crawford, 2009). Despite numerous studies documented the enabler effects of transport infrastructures, little research focuses on identifying their barrier effects (de Gruyter and Currie, 2016;Lara and Rodrigues da Silva, 2019;van Eldijk et al., 2022;Woodcock and Martin, 2016). This is particularly true in case of wider barrier effects on urban development.
Transport infrastructure can cause three types of barrier effects: a) direct, b) indirect, and c) catalytic/wider (van Eldijk et al., 2022). Direct barrier effects refer to an acceptable level of the barrier. Additional effort is required to overcome this type of barrier, but it does not necessarily influence people to change behaviour. For example, a roadrail level crossing may impose some congestion delay, but an individual may be willing to absorb such delay and take their usual travel routes rather than switching to a different route/mode. Indirect barrier effects result in changes in behaviour. For example, if the congestion level caused by a road-rail level crossing goes beyond the tolerable limit, an individual may change their usual travel route or travel mode. Wider barrier effects are more global, going beyond the behaviour that is linked with the severance itself. For example, changes in travel routes/ destinations may result in urban development occurring away from a crossing or reduce economic activities near a crossing. A range of examples of these three types of barriers have been documented by van Eldijk et al. (2022). They have also conceptualized the links among these barrier types and their five determinants as shown in Fig. 1. It outlines that the intensity of barriers increases with an increasing needs to cross the barriers, seperation of land uses, and the degree of separation posed by a transport infrastructure. In contrast, the intensity of barriers can be reduced through provisioning of alternative infrastructure to facilitate crossing the barriers and through increasing the ability of people to cross. However, these links have rarely been empirically tested in a systematic way, perhaps due to the difficulty of seprating barrier effects from other effects. Studies concerning the wider barrier effects of transport infrastructure in general and on local urban development specifically, are rare (Anciaes, 2013). Moreover, to date, studies have failed to establish causal relationships between barrier effects and urban development, perhaps due to a lack of relevant longitudinal data (de Gruyter and Currie, 2016;Lara and Rodrigues da Silva, 2019).
The barrier effects of transport infrastructure have been recognised as a policy concern in many countries across the world (Victoria Transport Policy Institute (BC) and Litman, 2000). The nature or type of transport infrastructure causing community severance can vary, such as the presence of highways/railways, elevated motorways, tunnels along with traffic volume, traffic speed, speed humps, lack of pedestrian crossing facilities, poor lighting/pavements/paths, or noise/air pollution (Anciaes et al., 2016;Taylor and Crawford, 2009). Various policy interventions have also been proposed to reduce the barrier effects, including the realignment of road networks or providing pedestrian crossing facilities (van Eldijk et al., 2022). However, the literature lacks a systematic investigation of the impacts of such interventions in terms of reducing barrier effects.
To-date, the assessment of community severance relies on ad-hoc procedures, mostly based on subjective evaluation, and often lacks to provide causal evidence, particularly in assessing the wider barrier removal effects (Anciaes et al., 2015). This study addresses this gap in the literature. It aims to identify the extent to which the removal of transport barriers unlocks the potential for local urban development. The study focuses on road-rail level crossing as a special case of community severance, and examines the consequence of replacing level crossings on surrounding land uses in the context of Melbourne, Australia. To our knowledge, this is the first study that attempts to assess the causal relationship between the removal of transport barrier and urban development.
A "level crossing" refers to an intersection where a railway line crosses a road or path and thereby acts as a barrier, creating delays for road users (pedestrians, cyclists, and motorists) to access destinations located on the other side of a railway line when the boom gates are in operation (Liang et al., 2020). When level crossing barriers are removed, congestion is reduced (State Government of Victoria Australia, 2022), perceived and real safety outcomes for road users are achieved (Anandarao and Martland, 1998;Mok and Savage, 2005;Ć irović and Pamučar, 2013), and travel becomes more attractive for both road and train users (van Eldijk et al., 2022). These are all direct barrier removal effects. These direct effects may act as an enabler to change peoples' behaviour resulting in indirect barrier removal effects. For example, reduced congestion attracts more traffic into the locality (Litman, 2017). Improved safety perception and accessibility encourage more people to walk and cycle to access railway stations and other destinations (Suarez-Balcazar et al., 2020). As outlined in Fig. 1, these indirect barrier removal effects are likely to flow on to wider barrier removal effects. For example, more pedestrians on the streets make a place more vibrant, which may attract more commercial activities and residential development (Quednau, 2018;Kamruzzaman et al., 2020). This discussion leads us to hypothesise that the wider level crossing barrier removal effects can be observed in the surrounding land use patterns. We tested this hypothesis using empirical data derived from an analysis of levelcrossing removals within the Melbourne metropolitan area.

Study context
Melbourne's Level Crossing Removal Project (LXRP) provided an opportunity to explore the research hypothesis. Melbourne is the capital of the state of Victoria, Australia, with approximately 5 million people. In 2016, the transport networks in Melbourne included 180 level crossings that were managed by boom gates to give priority to trains (Victorian Government, 2022). As would be expected, these level crossings slow down road traffic and cause driver delays. According to the 2019 Australian Infrastructure Audit, the combined cost of traffic congestion caused by these level crossings was expected to reach $10.4 billion by 2031, which could have increased as the city's train system was undergoing modernisation with the introduction of longer and more frequent trains (Melbourne Level Crossings Removal, 2022 showed that potentially fatal collisions, and signal faults negatively affected the safety, efficiency, and reliability of the transport network in Melbourne (Anandarao and Martland, 1998;Mok and Savage, 2005;Ć irović and Pamučar, 2013).
In response, the Victorian Government started implementing the LXRPs with an intention to replace 85 level crossings across metropolitan Melbourne by 2025 (Fowler,2021). The LXRPs in Melbourne is viewed as one of the most important public transport infrastructure projects in Australia and is the largest level crossing removal project in the world (Woodcock and Stone, 2016). At the time of writing, 65 level crossings had been removed.

Data and methods
Of the completed projects, 13 sites were selected for this research because the corresponding level crossing had been removed at least two years ago (Fig. 2). This enabled time for urban development processes to occur and meaningful changes in land use patterns to be measured. The selected sites all have a train station close by.
For each of the 13 LXRP project sites, one control site was identified to determine whether the changes in land use patterns had occurred as a result of wider economic effects (or chance) or if they were caused by the LXRP. Appendix A lists all the case and control sites. The control sites were selected using the following four criteria: a. Presence of a railroad-level crossing b. Similar distance from the central business district (CBD) to a corresponding case site c. Similar population density to the corresponding case site in 2015 (the baseline for the study) d. Similar diversity of land uses to the corresponding case site in 2015.
Two time periods were examined: a baseline in 2015 (i.e., before the onset of the crossing removal project), and a follow-up in 2020. The study used three spatial scales (200 m, 400 m, and 800 m) to study the effects of these transport infrastructure renewal projects. The 800 m scale was chosen as the maximum distance to correspond with the commonly used neighbourhood size in the literature (e.g., 20-min neighbourhood, transit-oriented development) (O'Gorman and Dillon-Robinson, 2021; Djunaidi et al., 2022;Thornton et al., 2022). Some land use interventions were a part of the LXRPs (new open spaces, bicycle parth, and play areas). These were implemented underneath (in case of skyrails) or over (in case of a tunnel) a railway line in the LXRP sites. These are direct interventions and are not a result of barrier  removal effects. As a result, land uses under/over the railway line were excluded from all the analyses (10 m buffer on both sides of the railway corridor centreline).

Data collection and description
The land-use patterns of the 13 case sites and 13 control sites at baseline were extracted from historic Nearmap images (spatial resolution 30 cm) and recorded in a geographic information system (GIS) database. If land-use patterns were unclear from the Nearmap images, they were clarified using Google Street View and administrative data. The procedure was first performed for the year 2015 ( Fig. 3), enabling comparison with the follow-up period in 2020 (Fig. 4). The accuracy of information was further enhanced through a site visit to each of the case and control sites in 2020. Land use at each study site was classified into 7 categories: parking, commercial, industrial, residential, service, open space, and vacant land. Appendix B depicts the full classification system applied.

Data analysis
Given the case-control nature of the study design (comparison of the LXRP vs. control sites) with data from two different time periods, the difference-in-difference (DiD) modelling framework was applied to estimate the true effects of the LXRPs on surrounding land uses. This method is particularly useful in quasi-experimental designs that compare the changes in outcomes over time between different groups. The DiD allows for causal inference even when randomisation is not possible (Fredriksson and Oliveira, 2019;Schwerdt and Woessmann, 2020). This effect was achieved by first calculating the difference in land use patterns between before (2015) and after (2020) the level crossing removals. This within-group comparison over time allows controlling for factors that are constant over time for both groups (i.e. the trendfor example, both case and control groups were subject to similar wider economic impacts). In the second step, the average difference (gain/ loss) in land use patterns of the control group was subtracted from the average difference (gain/loss) of the case group. This step allows us to compute the LXRP intervention effect on land use patterns. Mathematically, the DiD model took the following form: where, Yit is the dependent variable measured by the proportion of different land uses in case or control sites (i) at different time periods (t), β0 represents an intercept, β1 is the estimation term for the X1, which is the time factor (i.e., before/after), β2 is the estimation term for X2, which is the site factor (LXRP/control), β3 is the estimate for the interaction term (i.e., the difference-in-differences) between the X1 and X2. All statistical tests were performed in R (4.0.3). Land use patterns were measured by the sizes (area) of different land uses, which were then standardised to percentages. During the statistical analyses, the assumption of homoscedasticity was inspected by plotting the studentised residuals (i.e., division of a residual by an estimate of its standard deviation) versus the predicted values. The DiD analysis was performed separately for each spatial scale (200 m, 400 m, and 800 m) and each of the seven land use types, yielding a total of 21 model estimates. Table 1 shows the distribution of different land uses in the case and control sites in 2015 at the three spatial scales considered in this study. Clearly, residential and commercial are the most dominant pattern of land uses in both case and control areas. It is also evident from Table 1 that the proportion of residential land increases with the increase of spatial extent. This pattern is also evident for open spaces and service area. In contrast, the proportion of commercial land uses decreases as the spatial extent of analysis increases. A similar trend was also observed for the parking spaces. These findings suggest that commercial land use in both case and control sites was concentrated nearby the level crossing removal sites vis-e-vis the railway stations. As a result, a greater proportion of parking land use within the 200 m spatial scale is expected for two reasons: a) to provide parking facilities for the customers to access commercial land uses; and b) to provide park-and-ride facilities for the people taking the train. As the distance from the railway station increases, the study areas became more homogenous with residential development, which was supported by necessary open spaces and other services.

Distribution of different land uses
A comparison of different land uses between the case and control sites shows that both groups had an almost equal proportion of commercial (~25% at 200 m) and parking (~9%) land uses at the baseline in 2015 ( Table 1). The distribution of residential land use was marginally higher (50%) in the case sites compared to the control sites (45%) within the 200 m of the sites in 2015. The control sites, however, had a higher proportion of industrial land (11%) compared to the case sites (3%) at baseline. Overall, these finding suggest that the two groups were reasonably matched at the baseline. The slight variations in land use patterns between the case and control sites would not be a deterrant to assess the effects of the LXRP interventions if the parallel trend assumption (i.e. two groups would change with an equal rate without the interventions) of the estimated DiD model holds. We expect that this assumption holds due to the selection procedure applied for the control sites as noted in Section 3. Table 2 summarises the proportion of different land use in both case and control sites and their within group differences between 2015 and 2020 at the 200 m spatial scale. It is apparent that residential land use was the dominant type in both case and control sites occupying about half of the total area. However, residential land use lost it's dominancy in 2020 in the case site and commercial became the most dominant land use. In contrast, residential land use maintained it's dominancy in the control sites. Open space gained a substaintial proportion of land in 2020 in the case site, however, the changes in open spaces were found to be minimal in the control sites. These findings overall suggest some dissimilarities in the transition of land use patterns over the study period between the case and control sites, indicating some effects of LXRP on urban development. This is variefied in the estimated DiD model.

The effects of rail-crossing removal on the changes in land use patterns across different spatial scales
A DiD model was estimated based on Eq. 1 to identify if the DiD scores are statistically significant. A statistically significant score suggests that the changes in land use patterns are not due to chance but due to the LXRP interventions. Table 2 indicates significant DiD scores for the open space, commercial and residential land uses. Specifically, there was a significant rise in the amount of open spaces and commercial land uses at the LXRP sites at this scale. However, there was a significant decrease in the amount of land used for residential purposes in the case sites at this scale. The rate of changes in other land use patterns were not found to be significantly different between case and control sites over the study period.
An important objective of this study is to test how far the effects of the LXRP interventions can be observed from the intervention sites. Table 3 shows the similar DiD scores, but focused on 400 m spatial scale. The most pronounced differences in the land use patterns between 2015 and 2020 appear to be in the proportion of commercial and residential land uses. For the LXRP sites, there was a 10% increase in the proportion of commercial land use relative to only 0.73% increase in the control sites. On the other hand, there was a large decrease in the proportion of residential land use (16%) in the LXRP sites compared to a 0.9%  At the 800 m spatial scale, as shown in Table 4, the most pronounced differences in the land use patterns between 2015 and 2020 appear to be in the proportion of commercial and residential land uses. Specifically, for the LXRP sites, there was a large increase in the proportion of commercial land use of 8.03% compared to a smaller increase of 0.41% for the control sites. On the other hand, there was a large decrease in the proportion of residential land use (9.74%) in the LXRP sites compared to a slight decrease of 0.67% in the control sites. Additionally, there was also a slight increase in the proportion of parking land use (0.77%) and open spaces (1.25%) in the LXRP sites, whereas these land uses were relatively unchanges in the control sites. Overall, these findings echoed findings presented at the 400 m spatial scale. However, the DiD estimate shows that none of these trends were significantly different between the case and control sites, except for the commercial land uses. Table 3 shows that the proportion of changes in commercial land uses were significantly (at the 0.1 level) higher in the LXRP sites. These findings reinforce the findings presented at the 400 m spatial scale that the effects of LXRP are faded away as the distance increases from the intervention sites.

Discussion
This study provides the first empirical evidence of the wider barrier removal effects on urban development using road-rail level crossing in Melbourne as a source of barrier. While previous research has mostly concentrated on the direct and indirect barrier effects of transport infrastructure, no studies to date have explored whether there are catalytic urban development effects (i.e., reduction in urban development near the crossing, increase in economic activities near the replaced crossing) when such transport infrastructure barriers are removed.
This study found that there was a significant increase in open space at the LXRP sites relative to control sites between 2015 and 2020 at within the 200 m from the LXRP sites. Open space includes green spaces, such as grass, trees or other vegetation, and playgrounds. This increase in open space in urban areas has many benefits (Addas and Alserayhi, 2020;Wang and Stevens, 2020), including an increased physical activity of residents in the surrounding areas (Durand et al., 2011). This helps to improve the overall health and well-being of the community. For example, enhanced public space resulting from urban renewal in Copenhagen, Denmark, led to an increased time spent in the area and increased physical activity in a sample of adolescent residents (Andersen et al., 2017). Note that barrier removal can drastically increase the accessibility of open spaces for local residents, particularly for those who may have previously had difficulty crossing the railroad, such as children, the elderly, and people with disabilities. This may in turn increase the popularity of open spaces, thus contributing to an increase of this land use type.
Some of the most significantly pronounced wider changes were found in the proportion of land area used for residential purposes. At 200 m distance, residential land use decreased significantly (− 28.46%), perhaps due to an increase in open spaces (+11.93%) and commercial areas (+17.98%). It is possible that barrier removal can also increase the value of land in the surrounding areas, making it more attractive for commercial and other non-residential uses. This can lead to the redevelopment of residential land into commercial or other uses, and this idea is supported by the increased proportion of commercial land across LXRP sites. Note, though, that the land use measurement in the current work is limited since it does not describe the density of the corresponding land uses. For example, while the area of residential land use may have decreased, this does not account for the potential increase in multistorey apartments. Barrier removal can attract new residents to the area, leading to an increase in population density. This can make it necessary to remove older, lower-density housing to make way for new, higher-density housing that can accommodate more residents. In fact, a comparison of census data between 2016 and 2021 showed that there was a 21% increase in dwellings at the LXRP sites relative to 19% at the control sites. This implies that the number of people living in the LXRP areas grew over these years. This suggests that there was indeed a density uplift, which can explain the reduction of residential land use and simultaneous increase in population size. Therefore, measuring the density, as well as area, of residential and other land uses should be a focus in future research to gain a more accurate insight into the catalytic effects of a LXPR strategy.
There was an increase in parking area around the LXRP sites at 400 m. This mirrors a review article by Ravazzoli and Torricelli (2017), who described a systematic conversion of urban space into parking space to accommodate motorists in European cities. Increased proportion of land used for parking is logical given the increased accessibility caused by railroad crossing upgrade, making it more attractive for businesses and other activities that rely on vehicle access. This can lead to the development of more parking spaces to accommodate the increased demand. Similarly, barrier removal can also create new opportunities for urban development, including the construction of new buildings, which may require more parking spaces.
In contrast, the proportion of land used for industrial purposes remained constant between the case and control sites. This is not surprising, given that industrial land use is typically located further away from urban centers and transportation hubs, so it may not be as affected by barrier removal projects that are focused on improving access in more densely populated areas. Industrial land use is typically zoned separately from other land uses, such as residential or commercial, so it may not be as affected by changes in land use policies or development trends that are focused on other types of land uses. Finally, industrial land use is less dependent on changes in accessibility and mobility, since it doesn't rely on the same level of access as residential or commercial land use.
Another significant and relatively consistent change was the increase in commercial land use at the LXRP sites at the 200 m and 800 m distances, consistent with the results of Lee et al. (2020) in London. Commercial land use refers to retail premises, offices, and various other services. An increase in this land use suggests that after the LXRP site was completed, the resultant free land was used to improve the commercial availability of services for local residents around the targeted train stations, which subsequently could provide employment and service opportunities for local residents.
There was a significant reduction in vacant land in the LXRP sites at the 400 m spatial scale when compared to control sites. One possibility is that the barrier removal project served as a catalyst for revitalization in the area, and this led to a reduction in vacant land in those sites but not in the control sites. The removal of the barrier may have had a direct impact on the land use and development in the area, making it more desirable and accessible for people to invest in and develop.
Although this study found the evidence of some land use changes at all spatial scales (200 m, 400 m, and 800 m), the changes were more pronounced and statistically significant at the 200 m scale. The statistical significance level reduced with increasing distance from the sites. Similarly, the types of land uses that were statistically significant gradually declined as the distance increasedthe difference-indifference scores were statistically significant for 3, 2, and 1 land use types respectively at the 200 m, 400 m and 800 m spatial scales. This suggests that the wider, catalytic effect of the barrier removal project may be more pronounced at the vicity of the sites, but may not be limited to a specific, nearby area only, but rather, this wider effect can extend as far as 800 m away from the upgrading site. This implies that the wider effects of transport infrastructure renewal on land use should be studied and considered at a range of spatial scales than focusing on only the surrounding areas.
It should also be acknowledged that the current work measured the impact of the LXRP at time intervals that varied from two to four years after completion at the 13 selected sites. This may limit the chances of observing significant land use change since development of business and residential sites can take longer than this time. Indeed, based on previous literature, it is conceivable that the planning and preparation of changes, even at relatively small sites, can be a lengthy process. For example, Hurst and West (2014) examined land use changes six years after the completion of a construction project, while other studies conducted their analysis of land use changes over 10, 20, or more years (Kasraian et al., 2016).
Finally, the extent of project-related changes could be somewhat masked due to the Covid-19 outbreak and pandemic-related restrictions in the normal functioning of communities. Specifically, it is possible that some possible catalytic effects were suspended during the lockdown periods and corresponding restrictions, which added to delays in construction work and in the opening of new businesses. Thus, further measurements at later dates would compensate for these short-term restrictions and should be performed in future studies.

Conclusion
This study has investigated the potential effects of nenewing level crossing barriers on the built environment, specifically in terms of catalytic effects on land use. This is the first time this topic is explored in relation to level crossing removal projects. The findings suggest that these catalytic effects should be taken into account when making future policy decisions on such projects. This is important as the initial planning for the LXRP focused on planned effects such as the number of level crossings removed, but did not consider the potential catalytic effects that may occur after the project's implementation.
In conclusion, it is important to study the wider, catalytic effects of level crossing renewal and the upgrade of transport infrastructure on land uses in local areas. This study has provided initial evidence that these effects can have a significant impact on the built environment, and should be taken into account when planning and implementing these projects. By understanding the potential catalytic effects, policymakers and urban planners can better anticipate and plan for changes in land use, which can lead to more sustainable and equitable outcomes for the affected communities. Furthermore, it is important to have a holistic approach when studying the effects of transport infrastructure renewal on the built environment, as it can help to ensure that the projects are not only functional but also that they can contribute to the overall wellbeing of the community.

Declaration of Competing Interest
none.

Data availability
The data that has been used is confidential.