**1. Introduction**

Smart, reliable and connected service has been a long-standing goal of transit agencies. Smart transit network and service design must consider the service connectivity to allow users to travel to spatially diverse destinations [1,2]. However, providing direct connectivity for all origin—destination pairs is simply infeasible and impractical. Smart transfer is becoming an essential component of the transit trip. The extra e ffort in making transfers e fficient and convenient has deemed to be necessary to expand service coverage and to provide competitive area-wide connectivity [1,2]. Ironically, transfer is recognised by travellers as a significant impeding factor that disrupts the transit travel experience and deters the use of transit [3–5]. The literature formulates the e ffect of transfer in terms of the extra travel time such as the additional walking time, waiting time and in-vehicle travel time, incurred during transfer [5–7]. Another type of transfer penalty encapsulates subjective and psychological factors based on preferences, attitudes and perceptions of transit users [1,3,8].

The literature formulates the e ffect of transfer in the scalar form such as the extra walking time, waiting time, in-vehicle travel time and monetary transfer cost incurred during transfer [5–7]. Another type of transfer penalty encapsulates subjective and psychological factors based on preferences, attitudes and perceptions of transit users [1,3,8].

This study builds on the hypothesis that transfer location (i.e., the relative location of the transfer point to the trip origin and destination) has an impact on the attractiveness of the transit trip involving the transfer for travellers. This variable is quantified by the level of deviation of the transfer point from the straight path from the trip origin to destination. The deviation could be interpreted as intrinsic factors that reflect subjective or psychological impedance imposed by the transfer location. Whereas the burden of transfer has been quantified in terms of travel time and/or cost in the literature, we propose a new "transfer location" variable to represent the deviation in the travel direction. We examine if transit users tend to disfavour the transfer locations that deviate from a direct path to the trip destination and test this hypothesis by incorporating this novel explanatory variable to represent its underlying effect on the travel mode choice. To control the distance e ffect, this study proposes a combination of transformation techniques to keep only the deviation of transfer points. Despite an extensive range of research on transfer, the current literature has neglected the potential implication of transfer location in the decision-making of mode choice.

In the present study, we aim to improve the smartness of transfer and the explanatory ability of mode choice models by incorporating the transfer location variable. The findings of this study could contribute to smart transport planning and designing of new transit routes and smart services to enhance the transfer quality and to a more realistic assessment of transit service accessibility and connectivity. The spatial distribution of transfer points is analysed by their relative location with respect to the destination point. This study presents a transformation approach to convert the actual coordinates of the transit journey itineraries (i.e., origin, destination and transfer points) on a two-dimensional homogeneous geocoordinate. This approach may be useful for transit route choice and further transfer location analysis.

### **2. Literature Review**

Transfer is an essential and inevitable component of the transit journey (a chain of trips). It allows passengers to reach more destinations by switching to di fferent routes and modes, hence enhances the smartness of the transfer. In major cities with a multimodal transit system, the role of the transfer is more prevalent. In an integrated transit system, the focus is to provide seamless transfers between di fferent trips in a journey [2,9]. Smart transfers at strategic locations improve transit connectivity and expand spatial coverage of transit systems [10]. Despite its essential role, transfers are often seen as a burden in using transit [11]. Inconvenient transfers deter the use of transit for potential transit users and reduce the satisfaction level of existing transit users, which ultimately leads to reduction in the ridership.

The conventional way of quantifying the inconvenience of transfer has been by incorporating it into a generalised cost term to account for the extra monetary costs, travel time and discomfort incurred during the service transferring [5,12,13]. Transfer penalty can be measured as an equivalence of the travel time or money saving by taking the ratio between the coe fficients of transfer variables and time or cost variables. This ratio shows how much further people are willing to travel (time without transfer) or how much they are willing to pay (cost), to save one transfer, demonstrating the time and money that must be saved in order to justify one transfer [3,5]. The literature suggests that out-of-vehicle travel time is perceived as more onerous than in-vehicle travel time by transit users when making transfers [14,15]. In practice, the general rule of thumb is that walking and wait times are valued twice as much as the in-vehicle travel time [7,12]. Wardman, et al. (2001) sugges<sup>t</sup> that bus users value the wait time about 1.2 times higher than the in-vehicle travel time and the walk time 1.6 times higher than the in-vehicle travel time [7]. Generally, the wait time during transfer is also valued higher than the walking time during transfer [12,16].

Operational factors such as service reliability, headways regularity, on-time performance and the availability of adequate information a ffect the quality of transfers [5,12,17]. Providing a guaranteed connection and a through ticket for transfer could significantly reduce the perceived penalty of transfers [7]. An empirical study conducted in Haifa, Israel demonstrated that waiving a transfer fee resulted in a significant increase in the transit ridership [6]. Another study conducted in metropolitan Los Angeles showed that the users' satisfaction with the transit service transfer has little to do with the physical characteristics of the facility, but service frequency and reliability have more impact [18]. A study by Currie and Loader found that the volume of transfers could significantly increase along a major transit route when the service headway is 10 min or shorter [2].

Physical environmental factors such as stop and station amenities may a ffect the smartness of transfer services. Guo and Wilson reported that transit users are more likely to use the transfer service if escalators are available at the transfer station to assist with changing of levels [11]. Providing amenities such as benches, shades, water fountains and rest rooms would increase the comfort and convenience of transit users while waiting and transferring [5,12]. Security and safety, such as the presence of security sta ffs and the actual crime rates within the transit facilities would a ffect the perception of transfer quality [19]. A case study of the London underground train found that the worst transfer locations were the stations with the largest and most complex transfer environments, and the best transfer locations perceived were those stations with simple transfer environments [20]. In the case of whether to take a transfer or walk a longer distance to a destination, Guo and Wilson found that the demand of transfer decreases if walking environments are improved [11]. For example, if wider sidewalks exist along the non-transfer path, transit users are less likely to use a transfer service.

In an integrated transit system, more research seeks to understand and minimise the real cost of transfer inconvenience [15]. Much e ffort has been devoted to understand and minimise the cost of di fferent time components (e.g., walk and wait time) during transfer, such as the timed transfer concept. This concept optimises the slack time between the arrival of incoming vehicles with the departure of outgoing vehicles [21–23]. Ceder et al. developed a synchronised timetable by maximising the number of simultaneous bus arrivals at transfer nodes [14]. Shih et al. employed the heuristic model for the design of a coordinated network with transfer centres [24]. Similarly, Ting and Schonfeld used a heuristic algorithm to optimise the headways and slack times jointly for all coordinated routes, as the optimised slack times vary with di fferent variables such as headways, vehicle arrival time variance and transfer volumes [25].

As much as minimising the transfer time is important, transit users could also consider the travel direction towards the transfer point. Conventionally, the inconvenience to transfer caused by transfer location is considered as an increase in transit travel time, in the scalar form. This concept is similar to the "angular cost" concept presented by Raveau et al. to measure the directness of a chosen transit route [26]. The conventional route choice models account for the service level of the route alternatives and the socioeconomic and demographic characteristics of users [27]. Raveau et al. found that transit users tend to penalise routes that deviate from a direct path to their destination [26]. The "angular cost" is measured as a function of sin- θ 2 , where θ is the angle formed between the direct path to the destination (*OD*) with the origin-transfer (*OT*) straight route, weighted by the Euclidean distance to transfer point (*d*).

### **3. Study Area and Data**

The city of Brisbane accounts for approximately 70% of the total daily weekday trips in South East Queensland [28]. Brisbane has an extensive transit network of bus, rail and ferry systems, covering more than 10,000 km2. The recent report by the Queensland Government revealed that from January to March 2016, 27.38 million trips were conducted by bus, followed by 12.21 million trips by train, 1.71 million trips by ferry and 1.93 million trips by tram [29]. Bus ridership consisted of more than 63% of total transit ridership. This shows that the bus is the dominant transit mode in Brisbane. The benefit of the bus, in comparison to the train, tram and ferry, is that it has the flexibility to access

almost all locations where a road network is present. The nature of buses travelling on existing road networks gives more feasibility of adapting to change, such as the addition of new bus routes to serve more destinations. These considerations have steered the scope of this research towards bus ridership in Brisbane.

Brisbane's bus network may be characterised as a typical radial structure where more than 66% of the bus services operating to the city centre [30]. There are many routes heading in the same direction with very minor variations and no feeder or trunk services are currently provided. The CBD is the central hub for the bus system, where three grade-separated bus only corridors (busways) provide high-speed, high-capacity services to regional centres.

This study relies on two main data sources. First is the smart card data (big data), which is used to develop the transfer map of bus users in the study area. The one-day "go-card" data of Brisbane (24 November 2014, Monday) was used for the mapping. The data encapsulates the entire Brisbane area. The go-card is an electronic ticket for use on transit services throughout the network and records travel data when a traveller touches on at the start of any trip stage, and touches o ff at the end of the trip stage. This dataset contains information such as go-card ID, date of service, route ID, service ID, direction (inbound or outbound), boarding time and alighting time, boarding stop ID and alighting stop ID, ticket type, journey ID and trip ID. If it is a transfer journey, it would have a consecutive trip ID for each trip stage with the identical journey ID. According to TransLink, a journey is defined as the set of trip stages taken under one fare basis, while a trip is a ride on a single transit vehicle. This study adopts the same convention for the terms "journey" and "trip".

The second dataset used in this research is the 2009 Southeast Queensland Household Travel Survey (SQHTS). This single cross-sectional survey provides information on daily travel behaviour of all members of participating households, from 20 April through 28 June 2009. This includes how and why they travel, at what time of day journeys are made and the average journey distance and duration [28]. Respondents were also asked to report a range of personal information (e.g., age, gender, individual income, driver's license, etc.), and household related information (e.g., household size, number of vehicles, etc.).

### **4. Transformation Mapping of Transfer Coordinate**

This section presents a transformation approach to project the transfer locations on a homogeneous coordinate to examine the spatial distribution pattern of transfers. Travellers could be guided by the geographical images formed in minds, rather than the external maps, especially when individuals are familiar with the settings. These mental constructions su ffice as the sole source of spatial information. At instances when individuals are unfamiliar with the surroundings and need to rely on an external map, they still have to transcribe the cartographic information into their minds before they can act on the information [31,32]. In both instances, it is the spatial images in minds that best explain travellers' spatial behaviours.

### *4.1. Processing for Single-Transfer Journey Itineraries*

The smart card data was processed to filter out direct bus journeys and the journeys with two or more service transfers. The single transfer bus journeys account for about 20% of the total bus journeys. The journeys with two or more transfers are negligible less than 1% of the total bus journeys. The analytical framework of the transfer impact in this paper was developed applicable to only single-transfer journeys. The first step of the transformation mapping is to reconstruct travel itineraries by combining related trips from each smart card holder to form complete journeys from origins to destinations, including transfers. The data processing to construct single-transfer journeys is shown in Figure 1.

**Figure 1.** Process to construct single-transfer travel journeys.

The process starts by filtering out noise data such as the incomplete data of origin or destination information. A threshold of 60-min time gap (from the time when travellers alight a stop, to their next boarding time) is applied to identify whether two transactions are connected as a transfer journey. A different threshold has been chosen differently in the literature, ranging from 30 to 90 min [33,34], or a set of thresholds for different transit modes [35]. The threshold of the 60-min time gap is recommended in accordance with Brisbane's transit authority, based on the observed transfer behaviour and transit service characteristics [29]. If the transit user stays at a place for more than 60 min before making the next trip, those two trips are counted as separate trips, rather than a continuous journey through a transfer.

The next process is to distinguish return trips from single-transfer journeys. Studies have shown that transit users are willing to walk on average 400 or 500 m to bus stops [36–39]. A maximum distance threshold of 1 km from origin and destination was used to distinguish single-transfer journeys from return trips. To illustrate, the first bus stop could be located 500 m to the left of the journey's origin (e.g., the residence) and the last bus stop could be located 500 m to the right of the journey's destination (e.g., the residence). If the first and last bus stops are located less than 1 km apart, for the purpose of this study, it was assumed to be a return trip. This study was only interested in single-transfer journeys, so if there was any journey that had more than one transfer, the whole journey was removed from the dataset. After the reconstruction process, a total of 10,083 journeys were identified.
