The influence of weather on local geographical ...

Journal of Transport Geography 54 (2016) 66–80

Contents lists available at ScienceDirect

Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo

The influence of weather on local geographical patterns of bus usage Sui Tao a,⁎, Jonathan Corcoran b, Mark Hickman a, Robert Stimson c a b c

School of Civil Engineering, The University of Queensland, 4072, Australia School of Geography, Planning and Environmental Management, The University of Queensland, 4072, Australia School of Geography, The University of Melbourne, Victoria 3010, Australia

a r t i c l e

i n f o

Article history: Received 20 October 2015 Received in revised form 28 April 2016 Accepted 20 May 2016 Available online 28 May 2016 Keywords: Public transport Weather Geographic patterns Smart card data

a b s t r a c t This paper broadens the research on weather and public transport usage by considering the micro dynamics of the effect that various weather conditions impose on micro geographic patterns of bus ridership in Brisbane, Australia. A smart card data set and detailed measurements of weather, allied with a suite of statistical and visual analytic techniques, are employed to capture the effect of weather on the local variations of bus ridership. While changes in weather conditions do not significantly affect bus ridership at the system level, some marked influence was found for rainfall, wind speed and relative humidity at a sub-system level. In addition, discernible variations of both the magnitude and direction of weather's effect were found at the sub-system level. Developing a more geographically detailed understanding of the effect of weather on public transport services serves as a critical first step towards establishing a more weather-resilient public transport system. This new understanding has the potential to contribute to an evidence base that can be used to proactively adjust public transport services in response to changes in weather conditions across different parts of the network. Further research is needed to assess how transferable our findings are to other public transport and climatic contexts. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Transport and weather are intrinsically linked. Motorised transport in particular has been highlighted as a key sector contributing to greenhouse gas emissions and global warming (Chapman, 2007), with some research and reports claiming that climate change might bring about marked changes in our future daily weather patterns in a variety of ways, including increases in the number of severe weather days (IPCC, 2014; Hansen et al., 2006). Transport is also an integral part of our everyday life and subject to the influence of variations in weather conditions (Böcker et al., 2012; Koetse and Rietveld, 2009). We already know that inclement weather such as heavy rain and snow aggravates traffic congestion and causes increases in the number of incidents with the effect of degrading the operational efficiency of transport systems (Al Hassan and Barker, 1999; Call, 2011; Rakha et al., 2008). At an individual level, people's trip-making decisions have also been found to be influenced by weather conditions resulting in trip rescheduling, re-routing and cancellation (Cools et al., 2010; De Palma and Rochat, 1999; Sabir et al., 2010). It therefore follows that the operation and management of transport systems needs to consider daily variations of weather conditions to enhance resilience and operational

⁎ Corresponding author. E-mail addresses: [email protected] (S. Tao), [email protected] (J. Corcoran), [email protected] (M. Hickman), [email protected] (R. Stimson).

http://dx.doi.org/10.1016/j.jtrangeo.2016.05.009 0966-6923/© 2016 Elsevier Ltd. All rights reserved.

efficiency of transport systems in response (Koetse and Rietveld, 2009; Kim et al., 2013). Weather is capable of impacting public transport systems in a variety of ways including causing delays in delivering public transport services and making the experience of using public transport less comfortable (for example, waiting for a bus in the rain), therefore potentially reducing ridership (Changnon, 1996; Hine and Scott, 2000; Hofmann and O'Mahony, 2005). Understanding the influence of weather on public transport and particularly its impact on passengers' tripmaking behaviours therefore has potential to help improve public transport services by better meeting passengers' travel needs under different weather conditions (Arana et al., 2014; Guo et al., 2007; Kalkstein et al., 2009). Achieving this goal is of particular value given that many cities around the world have shown increasing interests to promote public transport use in order to pursue an overall more sustainable transport system (Cervero, 1998; Banister, 2011; Currie and Wallis, 2008). While some research has focused on developing an empirical understanding of the weather-public transport usage relationships, few studies have examined the geographic dimension. The scarcity of research in this area arguably calls for more attention especially given the strong spatial variations of both weather conditions (Hidalgo et al., 2008; Stewart and Oke, 2012; Theeuwes et al., 2014) and patterns of people's daily trip-making behaviours (Wang and Khattak, 2011; Tribby and Zandbergen, 2012; Salonen and Toivonen, 2013) and their interplay across a metropolitan area. As such it follows that weather conditions

S. Tao et al. / Journal of Transport Geography 54 (2016) 66–80

exert some notable influence on public transport and this influence varies across space. Thus, capturing the micro geographic patterns of public transport use in relation to variations in weather conditions has potential implications for prioritising public transport operation and management, for example, adjusting service frequency during a prolonged period of heavy rain. This study redresses the identified gap through an empirical examination of the geographic patterns of weather influencing public transport ridership. We take the bus network and its use in metropolitan Brisbane, Australia, as a case study. Drawing on a large smart card dataset and detailed measurements of weather conditions, we pay attention to three particular questions: • Do variations in weather conditions bring about shifts in bus ridership? • Is the effect of weather conditions on bus ridership homogeneous across the bus network? • Do variations in weather conditions create geographically discernible patterns of ridership within the bus network?

These questions are investigated using a suite of statistical and spatial analytical techniques to produce outputs that might inform the planning and operation of a city's bus services. The rest of the paper is structured as follows: Section 2 provides the theoretical background before introducing the study context and data sources in Section 3. Section 4 presents the analytic methods and results. Section 5 discusses the implications of the results for public transport management and policy in conjunction with identifying avenues for future research before drawing a set of tentative conclusions. 2. Theoretical background By combining theories in transport geography (e.g., time geography by Hägerstrand (1970)) and socio-psychology (e.g., the theory of planned behaviour by Ajzen (1991)), Van Acker et al. (2010) proposed an integrated conceptual model that describes the decision hierarchy that underpins an individual's travel behaviour (e.g., driving or taking public transport to work). According to this model, an individual's travel behaviour is subject to both reasoned (e.g., attitudes, preferences) and unreasoned (e.g., habits, impulsiveness) influences, which sit at the centre of the decision hierarchy. Such psychological decision process is then situated within a broader social (relating to one's social network, household compositions) and spatial context (relating to a city's built environment and infrastructure characteristics) that offers certain opportunities but also imposes specific constrain on an individual's travel behaviour decisions (i.e., where, when and how to make a trip). With the model by Van Acker et al. (2010) in mind, weather has been shown to have the potential to interface with individuals' attitudinal factors as well as the broader social and spatial context that may affect their travel behaviours. First, weather was found to exert evident impact on the spatial context of travel behaviour. For example, some researchers have found that adverse weather such as snowfall resulted in heightened hazards and reduced road speed compared to non-snow days (Al Hassan and Barker, 1999; Rakha et al., 2008; Call, 2011), thus reducing the accessibility of certain destinations for travellers. Under such circumstances, changes in travel decisions including rescheduling, rerouting or cancelling trips were observed (De Palma and Rochat, 1999; Cools et al., 2010). Second, concerning attitudinal factors, scholarship has revealed that people exhibited less preference towards active transport options than other forms of transport (in particular, private vehicles and public transport such as bus and train) during wet periods (Aaheim and Hauge, 2005; Sabir et al., 2010). In addition, Khattak and De Palma (1997) revealed that car users were reluctant to switch to using public transport (in particular, bus) during raining periods given the potential of getting wet and feeling cold while waiting for buses.

67

Last, the influence of weather on travel behaviour has been found to differ across different socio-demographic groups. For example, in a Belgian study, Khattak and De Palma (1997) found that commuters with children were less likely to change their travel patterns under inclement weather due to their household responsibilities. Despite the relevance of weather to the various aspects of individuals' trip-making decisions and behaviours, only a limited number of studies have investigated its influence on people's use of public transport (Guo et al., 2007; Hofmann and O'Mahony, 2005). For example, in a Chicago-based study, Changnon (1996) revealed a slight decrease of ridership of mass transit (3–5% for bus and 2.1% for train) associated with summer rainfall during weekdays. Within the same study context, Guo et al. (2007) highlighted that bus trips on weekends were more affected by adverse weather than weekday trips; and rail trips in general were less affected by weather compared to bus trips. Arana et al. (2014) found that occasional bus users were more influenced by weather than more frequent users on weekends (in particular, Sundays). In another related study, Kalkstein et al. (2009) found that across three urban areas (Chicago, the San Francisco Bay Area and Northern New Jersey) with distinctive climates, increase of rail ridership was associated with dry, comfortable days, while the reverse was the case for moist, cool days. In addition to the scarcity of existing research investigating the effects of weather on public transport usage, the geographic dimension of this particular issue has remained largely unexplored. We argue the need for research in this space evolves from two main reasons. First, weather conditions such as temperature, wind, rainfall are known to be capable of forming microclimates wherein continuous variations in weather over space may be observed within a metropolitan area (Hidalgo et al., 2008; Stewart and Oke, 2012; Theeuwes et al., 2014). Given such spatial heterogeneity (and in this regard, association), weather arguably has the potential to influence both the infrastructure and the level of service to various degrees across different parts of a public transport network, hence having spatially varying influence on ridership. While not focusing on the conventional public transport modes (e.g., bus, rail), the research by Helbich et al. (2014) and Corcoran et al. (2014) partially supports this point by highlighting some discernible differences of people's use of bicycles (and public bicycle in the latter case) under weather conditions across the metropolitan areas of Greater Rotterdam and Brisbane respectively. Second, people's travel behaviour has an intrinsic geographic component. This point has been increasingly recognised and affirmed by studies that have revealed collectively distinct patterns concerning peoples' trip-making decisions and behaviours. For example, empirical studies have found systematic variations of peoples' travel behaviours such as travel distance (Morency et al., 2011), time (Salonen and Toivonen, 2013), attitudes towards public transport (Páez, 2013), and use of transport information (Bagley and Mokhtarian, 2002; Wang and Khattak, 2011) over a city space. Factors including residential built environment (e.g., the diversity and design of a community), existing infrastructure (e.g., availability of public transport stops and services) and socio-economic conditions (e.g., income level, access to private cars) were identified as the underpinning influences. These studies have highlighted the importance of investigations at a range of spatial scales to reach a more detailed and comprehensive understanding of peoples' travel behaviour. Taken together the spatially heterogeneous dynamics of both weather and people's trip-making behaviours will induce geographically discernible patterns of public transport ridership. We aim to empirically test this assertion through analyses of weather-ridership relationships across a range of different spatial scales: (1) system level; (2) sub-system level; and (3) visually explore route level patterns. We justify such a geographically-based methodology given its capacity to test for the existence of geographically varying weather-ridership relationships and highlight parts of the public transport network that are particularly impacted by changes in weather.

68


3. Study context and data 3.1. Study area Brisbane, the third most populous Australian city, is the empirical setting for this study (Fig. 1). At its core is the Brisbane City Council local authority which has a population of a little over one million. Concerning Brisbane's transport mode shares, private cars account for over 85% of all daily trips, whereas the remainder is made by public transport, and to a lesser extent, active transport (BITRE, 2014a). While both rail and bus constituted the backbone components of Brisbane's transit network prior to 2000 (Rathwell and Schijns, 2002), the bus mode has become more prominent for transit users' tripmaking in the past 15 years due to a series of infrastructure upgrades, in particular the introduction of an exclusive busway (Hoffman, 2008; Mees and Dodson, 2011). According to the latest data, Brisbane's bus

network accounts for approximately 5.1% of all commuting trips versus 3.9% by rail transit (ABS, 2013). Given past and projected investment in the bus network, and particularly Brisbane's busway system (Department of Transport and Main Roads, 2011), buses are expected to play an increasingly important role in fulfilling people's travel needs across our study region (BITRE, 2014b). The advantages of using Brisbane as the study context are twofold. First is its subtropical climate regime that means the summer season (from December to February) is subject to storms characterised by spells of intense rainfall backed by relatively high temperatures (e.g., over 25 °C) and humidity (e.g., over 70%) (Bureau of Meteorology, 2015). This climatic context offers the possibility to examine dramatic shift in weather conditions and their potential to influence public transport ridership. Second, Brisbane is largely a monocentric city wherein the CBD contains close to 20% of all jobs and in excess of 40% of its population resides within 10 km from the city centre (ABS, 2015).

Fig. 1. Brisbane, the study context.


This results in a commuting pattern that is strongly CBD-centric (BITRE, 2013). Such a monocentric urban form provides a relatively simple context through which the weather-ridership associations can be examined in a more systematic manner than in polycentric settings. 3.2. Data sources To address our research questions and examine weather-ridership associations, there are two required data sets, namely, data on transit ridership and information on weather conditions. To examine this relationship at a range of different spatial scales it is desirable for individual transaction data describing public transport ridership allied with a meteorological database measuring weather conditions for multiple individual weather stations spread evenly across the study area. A search of available data sources identified two suitable data sets that met our requirements: (1) transit smart card transaction data from Translink (Brisbane's transit agency) covering a six-month period (November 2012 to April 2013); and (2) 30-minute measurements of weather conditions for 14 weather stations from the Australian Bureau of Meteorology (BOM). The smart card data records over 80% of all public transport trips made (including bus, train and ferry) in Brisbane. Each smart card record represents a single transaction, which is generated as a cardholder touches their smart cards when both boarding and alighting at a public transport stop. The information recorded for each smart card includes the date, route, direction (i.e., inbound for trips moving towards the city centre, and outbound for trips moving away from the CBD), smart card ID, boarding time and stop, and alighting time and stop. Route and direction information was not recorded for train and ferry trips and therefore these modes could not be included in this study. As such bus trips formed the sole focus. The BOM weather data includes the observations for four weather variables (temperature, rainfall, relatively humidity and wind speed). To generate a set of weather surfaces for the study region, an Empirical Bayesian Kriging (EBK) tool in ArcGIS was employed. This tool adopts an iterative Bayesian-based approach to produce robust geostatistical predictions (e.g., a surface of temperatures across the study area) using relatively sparse input data (Pilz and Spöck, 2007; Krivoruchko, 2012). This tool was particularly advantageous for our study given that meteorological data from 14 weather stations is relatively spatially dispersed. In addition to the four weather variables, apparent temperature, a composite thermal comfort index that incorporates temperature, relative humidity and wind speed (Bureau of Meteorology, 2010) was calculated to capture the collective influence of weather. The estimated weather values were then joined to the smart card data based on matching time of day and boarding and alighting stops. In addition the above two data sources, General Transit Feed Specification (GTFS), an open data source (Google Developers, 2012) was also employed. The GTFS contains detailed geographic (i.e., locations of all bus stops) and services (i.e., bus stops served by a given route) information of the bus network. Through associating matching stop IDs and route information in the smart card records, the GTFS was employed to geo-code bus stops and reconstruct route-level trips from smart card transaction data. 4. Data preparation and hypotheses Three interrelated empirical investigations form the basis of this study, namely, (1) system-wide, (2) sub-system and (3) route-level analysis. These analyses required a unique set of data pre-processing procedures, each of which is described next. 4.1. Data preparation for system-wide analysis Given we know that bus ridership varies strongly in accordance with calendar events (e.g., weekdays and public holidays), their influence

69

needs to be factored out to avoid confounding effects. In order to independently examine the effects of weather on ridership, we sampled the smart card data to remove systematic variations of ridership involving the following three steps. 1. Daily bus ridership (transaction count between the time period of 6:00 and 22:30) was first examined across weekdays, weekends and holidays. It was found that ridership on weekdays (83 days in total) was considerably higher than the other days (including 47 school holidays, 16 Saturdays, 16 Sundays and 10 public holidays), confirming the influence of calendar events. Given this, we selected only weekdays to form the focus of subsequent analyses. 2. Drawing on the remaining 83 weekdays, an Analysis of Variance (ANOVA) and t-tests were conducted across months and days. It was found that daily ridership in December (10 days) was significantly lower than other months, while higher in March and late April (26 days in total) due to the start of semesters for tertiary institutes. In addition, Fridays (17 days) were found to have significantly lower ridership compared to other weekdays. Given this, we excluded these days and retained the remaining 30 weekdays (17 days in November and 13 days in February). 3. Last, drawing on the remaining 30 weekdays we re-computed ANOVA and t-tests on the retained data to confirm the removal of systematic variations in ridership. Results revealed that no significant difference existed across months and days (i.e., Monday to Thursday). The outcome of this sampling resulted in 30 weekdays as the final sample for system-wide, sub-system and route-level analyses. To examine the potential influence of our sampling process on analyses, a visual (Fig. 2) and statistical examination (Table 1) was carried out to compare the weather conditions between the sampled and removed days. For both retained and removed data, relatively small variations were observed for temperature, wind speed and apparent temperature; and no significant difference was detected for the mean values of these weather variables (including relative humidity) between the two data sets. Notably higher mean and variability of rainfall were captured in the retained data, possibly due to the intensified storm season in February. This however was not deemed detrimental since we were able to capture some more extreme weather conditions in the sampled days. In summary, no major concerns were raised. 4.2. Data preparation for sub-system analysis Drawing on the data selected for system-wide analysis, we further pre-processed the retained data to enable sub-system analysis. This involved the following two steps: 1. To capture the ridership distribution across different parts of the bus network, we employed trip origin-destinations (OD) as the analytical units. Considering that the effect of weather is possibly of relevance whenever one is off-board (including transfer stages), a trip OD here is defined in compliance with a transaction record that involves one boarding and a subsequent alighting. Using three indicators, i.e., boarding stop, alighting stop and the network distance between OD (OD distance), over 80,000 unique OD pairs were identified. 2. Ridership over the sampled days was next calculated for each pair of the identified ODs. This forms the basis for the statistical tests of weather-ridership relationships at the sub-system level and their spatial distribution across the bus network. 4.3. Data preparation for route-level analysis Last, for route-level analysis, a recently developed geo-visual analytic technique, the flow-comap (Tao et al., 2014) was employed to explore passengers flow patterns under various weather conditions. We drew on the same data employed for both the system-wide and sub-system

70


Fig. 2. Weather conditions over the sample days.

analysis. Following Tao et al. (2014), three steps, namely, extracting transit service patterns using the GTFS data, reconstructing stop-tostop level trajectories and constructing flow matrices grouped by conditioning variables (e.g., direction of travel and time of day) are involved. To capture the incremental changes in a weather variable on the spatial patterns of bus ridership, the original method for computing flowcomaps was extended by including the following two data preprocessing steps: 1. For each weather variable of interest, the 30 weekdays were subsampled to remove extreme conditions of other weather variables. For example, when wind speed is the variable of interest, quartile ranges of the other controlled weather variables (e.g., temperature, wind and humidity) were calculated across the 30 weekdays. For each controlled variable, days within the first and forth quartile were removed. The same logic was applied in controlling for other weather variables (with N0.1 mm as the exclusion standard for rainfall as a controlled variable). 2. For each set of the sub-sampled data derived from the previous step (for example, 13 days of smart card data for wind speed), quartile or tertile ranges of the weather variable of interest (e.g., wind) were calculated to group the sub-sampled days into incremental weather categories (e.g., first, second and third quartiles of wind speed). Between different categories of weather conditions (e.g., first quartile versus second quartile of weed speed), standardised flow differences at a stop-to-stop level were calculated as the input for the weighted flow-comaps. 4.4. Hypotheses Scholarship has provided some evidence that variations in weather conditions profoundly influence people's daily activity patterns

(Horanont et al., 2013) and travel behaviour in particular (Böcker et al., 2012). Drawing on this body of work, we propose the following hypotheses. 1. Rain: Previous studies found considerable reductions in people's use of active and public transport during rainy days, e.g., Phung and Rose (2007); Tucker and Gilliland (2007) and Guo et al. (2007). Following this evidence, we expect there will be a strong negative association between rainfall and bus ridership. In addition, we expect that the effects of rainfall will be stronger on peripheral areas where less shelter is present. 2. Temperature: Previous studies found that in warmer climates (referring to temperatures of 28 °C and higher) exerted negative effects on outdoor activities and cyclists (Ahmed et al., 2010; Böcker et al., 2012). Following this evidence, we expect there will be a modest negative association between temperature and bus ridership given in our study context the daily average temperature was 23.8 °C backed with relatively small variations of ± 1.2 °C. While we note the potential effect of urban heat island in densely built-up areas, e.g., city centre (Theeuwes et al., 2014), we do not expect this to induce notable spatial variations of temperature-ridership relationship given such relatively mild temperatures. 3. Wind: Previous research captured the negative effects of strong wind (e.g., wind speed over 11 m/s) on travel behaviours including public transport use and cycling (Guo et al., 2007; Phung and Rose, 2007). However, given the persistence of light breezes over the sampled days (around 5 m/s), we expect there will be a moderate positive association between wind and bus ridership. We also expect the effect of wind will be stronger in the peripheral areas given the less shelter provided. 4. Humidity: While fewer studies have examined the effect of relative humidity on travel behaviour, one study, Miranda-Moreno and

Table 1 Comparisons of weather conditions between retained and removed data.

Retained data Removed data

Mean Std. deviation Mean Std. deviation

Temperature (°C)

Rainfall (mm)

Wind speed (m/s)

Relative humidity (%)

Apparent temperature

23.8 1.2 24.2 2.3

5 15.1 2.7 6.5

4.9 0.8 4.4 1.1

65.9 9.4 67.9 9.3

23 1.7 24.2 2.7


Nosal (2011), found that a humidity level of 60% coupled with high temperature (e.g., 28 °C) significantly reduced cycling behaviour. Following this and given an average humidity level of 66% in the study context, we expect there will be a moderate negative association between relative humidity and bus ridership. While humidity may be heighted by elevated heat in certain areas (e.g., the city centre) and potentially influence bus ridership, we do not expect such effect will be strong given the minor variability of temperatures in our study context.

5. Methods and results As discussed in Section 2, three empirical investigations were carried out at the system-wide, sub-system and route levels. The results of these investigations are presented in each of the following three subsections. 5.1. System-level associations Partial correlations were employed to capture the unique effect of each of the five weather variables on bus ridership (i.e., transaction count and passenger travel distance). In addition to daily ridership, ridership at different times of a day, i.e., morning (06:00–10:30), noon (10:00–14:30), afternoon (14:00–18:30) and evening (18:30– 22:30) was also examined due to their inherent difference in terms of travellers' trip purpose. Before computing partial correlations, ANOVA and t-tests were conducted for the ridership at four time periods (i.e., morning, noon, afternoon and evening) to detect any possible data bias. Significant differences were found for noon (between November and February) and evening (between Thursday and other weekdays), suggesting the existence of certain systematic changes in activity patterns. As such, partial correlations for daily, morning and afternoon periods are reported, which account for over 75% of all bus trips made (Table 2). No significant associations between weather and bus ridership were found with the exception of a positive correlation found between passenger travel distance and wind speed in the afternoon (partial correlation = 0.389, at the 95% level), suggesting stronger wind may encourage longer trip-making among bus passengers at that time period. However, as previously argued, such system-level results offer only a partial understanding of transit usage and weather, which may vary considerably across different parts of a transit network.

Table 2 Partial correlations between weather and bus ridership. Transaction count

Passenger travel distance

Partial cor.

Partial cor.

Daily Temperature Rainfall Wind speed Relative humidity Apparent temperature

−0.023 0.101 0.071 −0.003 0.008

−0.077 0.11 0.131 −0.059 0.064

Morning Temperature Rainfall Wind speed Relative humidity Apparent temperature

−0.184 0.123 0.223 −0.172 0.174

−0.137 0.139 0.181 −0.134 0.129

Afternoon Temperature Rainfall Wind speed Relative humidity Apparent temperature

−0.285 −0.046 0.331 −0.142 0.261

−0.349 −0.037 0.389⁎

Wherein: n is the sample size; and k is the number of variables controlled for; and r is the partial correlation. Examining mean values of the partial correlations (around 0 for all weather variables) and t-values (between 0.8 and 0.9 for all weather variables) across all trip ODs during morning and afternoon again indicated that on average, weather exerted a modest effect on bus ridership with no definitive direction. A scrutiny of the data shows that noticeable variations of the partial correlations (standard deviations of around 0.2) and t-values (standard deviations of 0.6 or more) exist across the trip ODs, indicating the spatial heterogeneity in terms of the changes in bus ridership in response to varying weather conditions. To further explore the sub-system variations of weather-ridership relationships and the associated spatial characteristics, we classified trip ODs into two groups based on t-values: those which were

Table 3 Comparing partial correlations across trip ODs. Group

Origin-based Temperature Rainfall Wind speed Relative humidity Apparent temperature Destination-based Temperature Rainfall Wind speed Relative humidity

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n−k−2 t¼r 1−r 2

ð1Þ

−0.208 0.325

⁎ p ≤ 0.05.

Morning

Apparent temperature

Afternoon

N

Mean partial cor. (abs)

Std. N deviation

Mean partial cor. (abs)

Std. deviation

1 2 1 2 1 2 1 2 1 2

688 10,831 954 10,565 737 10,782 760 10,759 674 10,845

0.446 0.144 0.465 0.147 0.444 0.145 0.447 0.144 0.445 0.143

0.058 0.097 0.082 0.099 0.056 0.098 0.058 0.097 0.058 0.096

686 11,417 820 11,283 715 11,388 757 11,346 728 11,375

0.445 0.144 0.463 0.143 0.447 0.145 0.457 0.145 0.445 0.145

0.059 0.097 0.075 0.097 0.061 0.098 0.068 0.098 0.056 0.098

1 2 1 2 1 2 1 2 1 2

753 10,766 923 10,596 730 10,789 785 10,734 703 10,816

0.448 0.145 0.466 0.147 0.448 0.146 0.452 0.145 0.449 0.144

0.060 0.097 0.081 0.099 0.059 0.098 0.060 0.097 0.059 0.097

721 11,382 826 11,277 721 11,382 752 11,351 755 11,348

0.446 0.143 0.462 0.142 0.445 0.145 0.450 0.145 0.446 0.144

0.059 0.097 0.074 0.096 0.060 0.097 0.061 0.098 0.059 0.097

5.2. Sub-system associations Next, we examined the associations of weather and bus ridership at the sub-system level and their spatial distributions. An examination of the ridership indicates that many trip ODs featured 30 passengers or less over the sampled days (or 1 passenger per day or less), for which any effect of weather observed would be difficult to justify and explain empirically. As such, we only reported the results for OD pairs with over 30 passengers across the sampled days (11,519 and 12,103 pairs for morning and afternoon periods respectively), which in total account for over 85% of the overall bus ridership. For each pair of the trip ODs, partial correlations were calculated between ridership (passenger count) and the weather conditions at the origin and destination stops separately (referred to as origin- and destination-based partial correlations hereafter). To capture the significance level of the partial correlations, t statistics were calculated that follow the form:

71

72


Table 4 Spatial characteristics of morning trip ODs, with rainfall as the variable of interest. Group Origin-based Origin location (metres)

Destination-based Destination location (metres)

Mean

Std. deviation

Mean difference

F

Sig

Group

1 2

7592 7429

5182 5005

163

7.111

0.008

Origin-based Origin location (metres)

1 2

3847 3732

4210 4384

114

3.881

0.049


influenced by a weather variable at the 95% level (with t ≥ 1.96, referred to as Group 1) and those which were not influenced by a weather variable (with t b 1.96, referred to as Group 2) (Table 3). The results identified a relatively small yet considerable amount of trip ODs as Group 1 under each weather variable. Rainfall appears to affect the largest number of trip OD pairs among the five weather variables, followed by relative humidity and wind. In comparison, fewer OD pairs were influenced by temperature and apparent temperature (especially for the morning period). A possible explanation is that passengers are acclimatised to travel in relatively warm weather (N20 °C) as a result of Brisbane's subtropical climate. A series of t-tests were next conducted to investigate spatial differences between Groups 1 and 2. Three spatial characteristics were used: origin and destination locations (calculated as the Euclidean distance from the CBD), and network distance between OD pairs (OD distance). Significant findings were detected for rainfall, wind speed and relative humidity in morning and rainfall in afternoon, while no significant results were obtained for temperature and apparent temperature. Concerning rainfall (Tables 4 and 5), for morning trips, Group 1 was found to be slightly more distant from the CBD than Group 2 in terms of trip origins and destinations (e.g., over 100 m), while the reverse is the case for afternoon trips. For wind speed (Table 6), trip ODs in Group 1 were found to be associated with destinations located farther away from the CBD than Group 2 (by over 300 m); and for relative humidity (Table 7), longer average OD distances were found for Group 1 compared to Group 2 (by 450–700 m). Building on the results presented in Tables 4-6, a local Moran's I (Anselin, 1995) was employed to explore the spatial distributions of trip origins and destinations for which ridership were significantly associated with weather across the study context (i.e., Group 1). This geostatistical technique is able to capture the local spatial association and clustering among spatial units. Local Moran's I here is defined as: X I ¼ zi wij z j ð2Þ j

Wherein: zi and zj are the origin- or destination-based partial correlations; and wij is the spatial weight matrix defined by inverse distance among trip origins or destinations. After calculating this local statistics, two major types of clusters were identified at the 95% confidence level for rainfall, wind and humidity: (1) clusters of trip origins (or destinations) for which ridership was

Table 5 Spatial characteristics of afternoon trip ODs, with rainfall as the variable of interest. Group Origin-based Origin location (metres)


Table 6 Spatial characteristics of morning trip ODs, with wind speed as the variable of interest.

Mean

Std. deviation

Mean difference

F

Sig

Mean

Std. deviation

Mean difference

F

Sig

1 2

4084 3718

4674 4348

366

9.9

0.002

1 2

4023 3722

4715 4345

300

12.518

0.000

positively associated with changes in weather; and (2) clusters of trip origins (or destinations) for which ridership was negatively associated with changes in weather. With rainfall as the variable of interest, Fig. 3 highlights a number of spatial clusters for morning and afternoon trip ODs. For the both periods, positive clusters were found to be associated more with trips with origins and destinations located within or surrounding the inner areas of Brisbane, while negative clusters appear to concentrate along trips originating from and bound for peripheral areas. These spatial patterns largely agree with our hypothesis that the peripheral areas of the study context are more vulnerable to the effects of rainfall, given the less shelter and sparser bus services provided. It also appears that some positive clusters were along or surrounding the busway, suggesting the possibility that the better shelters at the busway stations (Currie and Delbosc, 2010) encouraged bus use under rainy periods. This however needs confirmation through more detailed investigation as presented in the following section. For wind speed and relatively humidity (Fig. 4), however, fewer clusters were found and their spatial patterns are less discernible compared to rainfall, which also calls for further investigation. 5.3. Sub-system flow dynamics To extend the results from the previous section, the spatial patterns of passenger flows associated with the variations in weather conditions were visually explored using the flow-comap technique described in Section 4.3 (Figs. 5–8). To our knowledge, this study marks the first to apply a geo-visual analytic technique to examine weather-ridership associations at a micro geographic scale (i.e., a stop-to-stop route level). For simplicity of display, inbound and outbound trips as recorded in the smart card data were visualised separately. It is noted that the warmer the colour (e.g., red), the larger the increase of the ridership along that pathway is relative to other pathways of the network, and vice versa. Now we examine the effect of the three variables on the spatial patterns of ridership in sequence. First, considering the morning period (Fig. 5), light rainfall (i.e., the first of the three tertiles) appears to induce decreases in passenger flows along specific pathways, in particular for inbound trips. However, with higher levels of rainfall (particularly the third tertile), marked increases of passenger flow are observed along a number of pathways for both inbound and outbound trips. This is chiefly the case for busway-based trips which collectively experienced pronounced growth

Table 7 Spatial characteristics of morning trip ODs, with relative humidity as the variable of interest. Group

Mean Std. deviation Mean difference F

1 2

4003 4314

4248 4658

−310

26.367

0.000

Origin-based OD distance (metres) 1 8091 2 7635

1 2

7036 7180

4847 5081

−144

6.932

0.008

Destination-based OD distance (metres) 1 8321 2 7617

Sig

6046 5623

455

4.826 0.028

5938 5629

703

7.367 0.007


73

Fig. 3. Spatial clusters for partial correlations with rainfall as variable of interest.

in ridership during heavier rain, whereas some pathways in the far north and east saw marked reductions in passenger flow under the same weather conditions. While partially affirming our hypothesis regarding variations of rain-ridership relationships over space, such changes in bus ridership also suggest the possibility of a nonlinear relationship between rainfall and bus use along certain routes, and the existence of a rainfall threshold (partially captured by the third tertile of rainfall here) beyond which modal shifts were occurring. Here, given the elevated risk for road accident and reduced road speed during heavier rain, some car users might be less inclined to drive, and such inclement weather also discouraged walking and cycling (Sabir et al., 2010). As a result, some individuals might change to use buswaybased services for their trips.

The spatial patterns of bus ridership during the afternoon period (Fig. 6) appear to be similar to that of the morning period especially during the heavier rain (i.e., the third tertile of rainfall). In particular, for both inbound and outbound trips, notable increases in passenger flows were found to concentrate on the busway. Major decreases in passenger flows, on the other hand, occurred along trips bound for the outer west areas. The effects of wind speed on morning bus ridership appear to be mainly positive in that notable increases in passenger flows took place along a number of inbound pathways, and to a lesser extent, some outbound pathways towards locales in the north (Fig. 7). In line with our hypothesis, this might be attributed to that the onset of wind can reduce the sense of heat stress, which as such encourages bus use. Yet again,

74


Fig. 4. Spatial clusters of partial correlations with relative humidity and wind speed as variables of interest.

some local variations were identified concerning the effects of wind. In particular, some segments along the south busway experienced decreases in ridership in stronger wind. While it cannot be stated as exact, one possible explanation is that the shelters of some busway stations offered a higher level of protections from weather conditions, particularly wind and rain. Last, an examination of the effect of humidity on bus ridership (Fig. 8) appears not to highlight any highly discernible spatial patterns for either inbound or outbound trips. That said, a weak trend can still be identified concerning the effects of humidity that as humidity intensified to a higher level (from second to third tertile), more pathways around Brisbane experienced marked decrease in passenger flows while few saw increases in passenger flows. This pattern echoes our

hypothesis that highly humid weather was deemed less pleasant and discouraging for bus use given a subtropical climate. 6. Discussion and conclusions While it is increasingly recognised that changes in weather conditions have profound effect on people's use of public transport (Böcker et al., 2012; Guo et al., 2007), empirical work on this issue has been limited. In addition, few if any have paid attention to the geographic patterns of weather's effects on public transport usage, despite the potentially complex interplay between weather and people's travel behaviour over space. To redress this research gap, this study aims to capture local patterns of public transport use in association with weather


75

Fig. 5. Weighted flow-comap for morning trips conditioned on rainfall. 1 Tertile (T) is used to categorise level of rainfall into 3 groups. No rain = 0 mm over 4.5 h T1 = 0.016 to 0.1 mm over 4.5 h T2 = 0.036 to 0.363 mm over 4.5 h T3 = 0.363 to 3.95 mm over 4.5 h.

using a geographically focused approach. Drawing on transit smart card data in conjunction with detailed weather measurements, a suite of statistical and geo-visual techniques were for the first time employed to explore the effects of weather conditions on bus ridership at a range of spatial scales (i.e., system-wide, trip OD and route level). From these analyses, a series of novel insights as well as implications are drawn with the potential to inform the planning and operation of a bus network. Each is now discussed. First, the results of partial correlation analysis indicate while there was essentially no significant association between changes in weather and system-wide bus ridership over the sampled days, a small yet substantial subset of trip ODs were highlighted to be significantly

associated with weather variables particularly including rainfall, wind and humidity during morning and afternoon periods. Through a series of local statistical analysis, some discernible spatial patterns particularly for rainfall-ridership relationships were detected. Specifically, clusters of positive associations between rainfall and bus ridership were found for trip ODs located within or near the inner areas of Brisbane where more shelter might be present, while clusters of negative associations spatially coincide with trips moving towards peripheral areas wherein passengers might be more exposed to inclement weather. These findings provide complementary insights to the previous research that solely examined system-wide rain-ridership relationships (e.g., Hofmann and O'Mahony (2005); Kalkstein et al. (2009); Arana et al. (2014)) by

76


Fig. 6. Weighted flow-comap for afternoon trips conditioned on rainfall. 2 No rain = 0 mm over 4.5 h. T1 = 0.016 to 0.1 mm over 4.5 h T2 = 0.1 to 1.4 mm over 4.5 h T3 = 1.3 to 19.3 mm over 4.5 h.

highlighting that both the magnitude and the direction of the effects of weather on public transport passengers may significantly vary across space. To further locate where the most marked changes in ridership occurred, we next employed the flow-comap to visually examine the sub-system flow dynamics in associations with variations in rainfall, wind and humidity. Some more detailed patterns were revealed. First, compared to relatively light rain, heavier rain appeared to markedly prompt bus ridership along the busway, implying the existence of a threshold of rainfall that might trigger demand increase for buswaybased services potentially from car users as well as cyclists and pedestrians. Concerning wind, increases in ridership were however observed

mostly along pathways originating from relatively remote and less dense locales, wherein the cooling effect of wind on heat stress might be more prominent. Last, it was found that high humidity mainly reduced bus use across the study context, suggesting its role as a calming factor for bus use within a subtropical climatic context. Some recommendations for bus operation and planning can be derived, which are mainly threefold. First, the spatial patterns revealed concerning weather and bus ridership can be used as a basis for upgrading the infrastructure of bus network. For example, better weather shelters may be added to the bus stops within areas where negative clusters for rainfall were identified. Second, drawing on the sub-system flow patterns, it is possible to use these to inform a series


77

Fig. 7. Weighted flow-comap for morning trips conditioned on wind speed. 3 Quartile (Q) is used to categorise level of wind speed into 4 groups. Q1 = 2.22 to 2.52 m/s over 4.5 h Q2 = 2.44 to 2.84 m/s over 4.5 h Q3 = 2.64 to 4.14 m/s over 4.5 h Q4 = 4.13 to 5.16 m/s over 4.5 h.

of weather-based re-allocations and adjustments to bus schedules particularly in response to rainfall. Within the study context, such service adjustment might include supplementing services operating along the busway (refer to Fig. 1) during prolonged periods of heavy rain (up to 4 to 20 mm over 4 h here) in both morning and afternoon. Prior to making such service adjustment, there is need to supplement this evidence with additional information that suggests the existing service capacity is exceeded. Given the operational constraints of a bus network, the re-allocation of bus services may be achieved through re-directing vehicles from geographically proximate routes (in the first instance), particularly those that are more likely to experience reduced ridership during inclement weather. However, we note that care must be taken

when adopting such a strategy. In particular, a minimum level of service should be maintained when relocating bus services from certain routes such that the travel needs of more captive passengers (e.g., lower income, student passengers) are still met. As such there is a need to undertake a comprehensive population-based assessment of travel demands prior to making any ad hoc change of the bus services to ensure the adjustments adequately meet equity considerations. Last, the analytical framework developed and applied here can serve as a platform which the investigation of the weather-ridership relationship within other public transport and climatic contexts may draw upon to assess their geographical variations prior to setting up a more context specific operational plan for public transport services.

78


Fig. 8. Weighted flow-comap for morning trips conditioned on relatively humidity. 4 Quartile (Q) is used to categorise level of relative humidity into 4 groups. Q1 = 52 to 65% over 4.5 h Q2 = 65 to 66% over 4.5 h Q3 = 66 to 71% over 4.5 h Q4 = 69 to 76% over 4.5 h.

There remain four areas that could form avenues for future research. First, to avoid the confounding effects of calendar events, we truncated a six-month set of smart card and focused only on certain weekdays. Future research may examine the geographical patterns of the weather influencing bus usage across different calendar events (e.g., other weekdays, weekends and school holidays). This will require a larger smart card data set (e.g., a minimum of two years as in the case of Arana et al., 2014) to capture enough samples for each calendar event. Second, the primary goal of this study was to detect the existence and patterns of weather-ridership associations at a local level. Given such exploratory purpose, relatively simple method (partial correlation) was applied. To build on the findings of this study, future work should target the testing

of the weather-transit associations in a modelling exercise to establish causality. This is certainly not a trivial exercise and would entail employing a spatial modelling framework capable of capturing the interaction of local shifts in weather patterns relative to ridership: geographically weighted panel regression would seem one appropriate candidate for this modelling exercise (Yu, 2010). Third, previous research found that train services were less affected by adverse weather (e.g., rainy days) than bus, possibly due to their superior service attributes including higher service frequency and better sheltered stations (Guo et al., 2007). Given this, it would be of value to investigate whether changes in weather exert global and more importantly, local modal shifts in ridership. However, this type of investigation is not possible


here due to data limitations previously noted. Last, for future research, efforts should be focused on investigating the individual behavioural responses of transit passengers and how these are conditioned by their associated socio-demographic characteristics under different weather conditions. This will enable a more detailed understanding of certain behavioural changes (e.g., cancelling a particular trip and a modal shift) made by transit passengers across different socio-demographic groups (e.g., workers versus tertiary students). To achieve this would require the acquisition of additional personal information (e.g., socioeconomic characteristics and type of card holder-student, adult and senior citizen) from smart card holders (see Utsunomiya et al. (2006) for an example). Such a set of data is not currently available for our research. To conclude, we have contributed to the understanding of the relationship between weather and public transport ridership by shedding light on their local geographical patterns. Findings herald a series of operational and strategic implications that, if drawn upon as a new evidence base, have the potential to enhance transit operations under different weather conditions. Considering the predicted increase in the number and magnitude of extreme weather conditions that might be induced by climate change, there is a compelling need to deepen our understanding of the effect of weather as a situational factor influencing people's travel behaviour. As such, this study may stimulate future research in this area to help establish more weather resilient public transport systems. Acknowledgements We would like to thank Ms. Hui Xiao for her help in the coding process. We also acknowledge the valuable input of two anonymous reviewers and the Editor Prof. Kevin O′Connor. Finally, we would like to thank Translink for kindly providing the smart card transaction data. The views expressed in the paper are solely from the authors and do not necessarily reflect those of Translink. References Aaheim, H.A., Hauge, K.E., 2005. Impacts of climate change on travel habits: a national assessment based on individual choices. CICERO Report. ABS, 2013. 2011 Census QuickStats, [Online]. ABS, Canberra (Available: http://www. censusdata.abs.gov.au/census_services/getproduct/census/2011/quickstat/0? opendocument&navpos=220 [Accessed 18th, November 2013]). ABS, 2015. TableBuilder [Online]. Australian Bureau of Statistics, Canberra, Australia (Available: http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder? opendocument&navpos=240 [Accessed 15th, February 2015]). Ahmed, F., Rose, G., Jacob, C., 2010. Impact of weather on commuter cyclist behaviour and implications for climate change adaptation. Australasian Transport Research Forum (ATRF), 33rd, 2010, Canberra. ACT, Australia (Year). Ajzen, I., 1991. The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50, 179–211. Al Hassan, Y., Barker, D.J., 1999. The impact of unseasonable or extreme weather on traffic activity within Lothian region, Scotland. J. Transp. Geogr. 7, 209–213. Anselin, L., 1995. Local indicators of spatial association—LISA. Geogr. Anal. 27, 93–115. Arana, P., Cabezudo, S., Peñalba, M., 2014. Influence of weather conditions on transit ridership: a statistical study using data from Smartcards. Transp. Res. A Policy Pract. 59, 1–12. Bagley, M.N., Mokhtarian, P.L., 2002. The impact of residential neighborhood type on travel behavior: a structural equations modeling approach. Ann. Reg. Sci. 36, 279–297. Banister, D., 2011. Cities, mobility and climate change. J. Transp. Geogr. 19, 1538–1546. BITRE 2013. Population growth, jobs growth and commuting flows in South East Queensland. In: Transport, D. o. I. a. (Ed.). (Canberra, Australia). BITRE 2014a. Urban public transport: updated trends. In: BITRE (Ed.). (Canberra). BITRE 2014b. Public transport use in Australia's capital cities: modelling and forecasting. In: BITRE (Ed.). (Canberra, Australia). Böcker, L., Dijst, M., Prillwitz, J., 2012. Impact of everyday weather on individual daily travel behaviours in perspective: a literature review. Transp. Rev. 33, 71–91. Bureau of Meteorology, 2010. Thermal Comfort Observations [Online]. Australian Government (Available: http://www.bom.gov.au/info/thermal_stress/ [Accessed 7th, April 2015]). Bureau of Meteorology, 2015. Brisbane Metro in 2013: Warm Year, Average Rainfall [Online]. Australian Government, Canberra (Available: http://www.bom.gov.au/ climate/current/annual/qld/archive/2013.brisbane.shtml#summary [Accessed 8th, September 2015]). Call, D.A., 2011. The effect of snow on traffic counts in western New York State. Weather Clim. Soc. 3, 71–75.

79

Cervero, R., 1998. The Transit Metropolis: A Global Inquiry. Island Press, Washington, DC. Changnon, S., 1996. Effects of summer precipitation on urban transportation. Clim. Chang. 32, 481–494. Chapman, L., 2007. Transport and climate change: a review. J. Transp. Geogr. 15, 354–367. Cools, M., Moons, E., Creemers, L., Wets, G., 2010. Changes in travel behavior in response to weather conditions. Transp. Res. Rec. J. Transp. Res. Board 2157, 22–28. Corcoran, J., Li, T., Rohde, D., Charles-Edwards, E., Mateo-Babiano, D., 2014. Spatiotemporal patterns of a public bicycle sharing program: the effect of weather and calendar events. J. Transp. Geogr. 41, 292–305. Currie, G., Delbosc, A., 2010. Bus rapid transit in Australasia: an update on progress. Built Environ. 36, 328–343. Currie, G., Wallis, I., 2008. Effective ways to grow urban bus markets – a synthesis of evidence. J. Transp. Geogr. 16, 419–429. De Palma, A., Rochat, D., 1999. Understanding individual travel decisions: results from a commuters survey in Geneva. Transportation 26, 263–281. Department of Transport and Main Roads, 2011. Connecting SEQ 2031 – An Integrated Regional Transport Plan for South East Queensland. Google Developers, 2012. Transit [Online]. Available: https://developers.google.com/ transit/gtfs/ [Accessed April 2013]. Guo, Z., Wilson, N., Rahbee, A., 2007. Impact of weather on transit ridership in Chicago, Illinois. Transp. Res. Rec. J. Transp. Res. Board 2034, 3–10. Hägerstrand, T., 1970. What about people in regional science? Pap. Reg. Sci. 24, 7–24. Hansen, J., Sato, M., Ruedy, R., Lo, K., Lea, D.W., Medina-Elizade, M., 2006. Global temperature change. Proc. Natl. Acad. Sci. 103, 14288–14293. Helbich, M., Böcker, L., Dijst, M., 2014. Geographic heterogeneity in cycling under various weather conditions: evidence from Greater Rotterdam. J. Transp. Geogr. 38, 38–47. Hidalgo, J., Pigeon, G., Masson, V., 2008. Urban-breeze circulation during the CAPITOUL experiment: observational data analysis approach. Meteorog. Atmos. Phys. 102, 223–241. Hine, J., Scott, J., 2000. Seamless, accessible travel: users' views of the public transport journey and interchange. Transp. Policy 7, 217–226. Hoffman, A., 2008. Advanced network planning for bus rapid transit: the “Quickway” model as a modal alternative to “Light Rail Lite”. TRIS. Hofmann, M., O'Mahony, M., 2005. The impact of adverse weather conditions on urban bus performance measures. Intelligent Transportation Systems, 2005. Proceedings. 2005 IEEE, 13–15 Sept. 2005, pp. 84–89 (Year). Horanont, T., Phithakkitnukoon, S., Leong, T.W., Sekimoto, Y., Shibasaki, R., 2013. Weather effects on the patterns of people's everyday activities: a study using GPS traces of mobile phone users. PLoS One 8, e81153. IPCC, 2014. Climate Change 2014: Synthesis Report. In: Core Writing Team, Pachauri, R.K., Meyer, L.A. (Eds.), Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. IPCC, Geneva, Switzerland. Kalkstein, A.J., Kuby, M., Gerrity, D., Clancy, J.J., 2009. An analysis of air mass effects on rail ridership in three US cities. J. Transp. Geogr. 17, 198–207. Khattak, A.J., De Palma, A., 1997. The impact of adverse weather conditions on the propensity to change travel decisions: a survey of Brussels commuters. Transp. Res. A Policy Pract. 31, 181–203. Kim, J., Mahmassani, H.S., Alfelor, R., Chen, Y., Hou, T., Jiang, L., Saberi, M., Verbas, Ö., Zockaie, A., 2013. Implementation and evaluation of weather-responsive traffic management strategies: insight from different networks. Transp. Res. Rec. J. Transp. Res. Board 93–106. Koetse, M.J., Rietveld, P., 2009. The impact of climate change and weather on transport: an overview of empirical findings. Transp. Res. Part D: Transp. Environ. 14, 205–221. Krivoruchko, K., 2012. Empirical Bayesian Kriging implemented in ArcGIS geostatistical analyst. ArcUser, the Magazine for Esri Software Users. ESRI, 10, p. 6. Mees, P., Dodson, J., 2011. Public transport network planning in Australia: assessing current practice in Australia's five largest cities. Res. Pap. 34. Miranda-Moreno, L., Nosal, T., 2011. Weather or not to cycle. Transp. Res. Rec. J. Transp. Res. Board 2247, 42–52. Morency, C., Páez, A., Roorda, M.J., Mercado, R., Farber, S., 2011. Distance traveled in three Canadian cities: spatial analysis from the perspective of vulnerable population segments. J. Transp. Geogr. 19, 39–50. Páez, A., 2013. Mapping travelers' attitudes: does space matter? J. Transp. Geogr. 26, 117–125. Phung, J., Rose, G., 2007. Temporal variations in usage of Melbourne's bike paths. Proceedings of 30th Australasian Transport Research Forum. Forum Papers, Melbourne, p. 1CD (Year). Pilz, J., Spöck, G., 2007. Why do we need and how should we implement Bayesian Kriging methods. Stoch. Env. Res. Risk A. 22, 621–632. Rakha, H., Farzaneh, M., Arafeh, M., Sterzin, E., 2008. Inclement weather impacts on freeway traffic stream behavior. Transp. Res. Rec. J. Transp. Res. Board 2071, 8–18. Rathwell, S., Schijns, S., 2002. Ottawa and Brisbane: comparing a mature busway system with its state-of-the-art progeny. J. Public Transp. 5, 163–182. Sabir, M., Ommeren, J., Koetse, M.J., Rietveld, P., 2010. Impact of weather on daily travel demand. Proceedings of the Tinbergen Institute Discussion Paper. VU University, Amsterdam (Retrieved from:). Salonen, M., Toivonen, T., 2013. Modelling travel time in urban networks: comparable measures for private car and public transport. J. Transp. Geogr. 31, 143–153. Stewart, I.D., Oke, T.R., 2012. Local climate zones for urban temperature studies. Bull. Am. Meteorol. Soc. 93, 1879–1900. Tao, S., Rohde, D., Corcoran, J., 2014. Examining the spatial–temporal dynamics of bus passenger travel behaviour using smart card data and the flow-comap. J. Transp. Geogr. 41, 21–36. Theeuwes, N.E., Steeneveld, G.J., Ronda, R.J., Heusinkveld, B.G., van Hove, L.W.A., Holtslag, A.A.M., 2014. Seasonal dependence of the urban heat island on the street canyon aspect ratio. Q. J. R. Meteorol. Soc. 140, 2197–2210.

80


Tribby, C.P., Zandbergen, P.A., 2012. High-resolution spatio-temporal modeling of public transit accessibility. Appl. Geogr. 34, 345–355. Tucker, P., Gilliland, J., 2007. The effect of season and weather on physical activity: a systematic review. Public Health 121, 909–922. Utsunomiya, M., Attanucci, J., Wilson, N., 2006. Potential uses of transit smart card registration and transaction data to improve transit planning. Transp. Res. Rec. J. Transp. Res. Board 1971, 119–126. Van Acker, V., Van Wee, B., Witlox, F., 2010. When transport geography meets social psychology: toward a conceptual model of travel behaviour. Transp. Rev. 30, 219–240.

Wang, X., Khattak, A., 2011. Role of travel information in supporting travel decision adaption: exploring spatial patterns. Transportmetrica A Transp. Sci. 9, 316–334. Yu, D., 2010. Exploring spatiotemporally varying regressed relationships: the geographically weighted panel regression analysis. Joint International Conference on Theory, Data Handling and Modelling in GeoSpatial Information Science. Hong Kong, 26–28 May 2010. ISPRS Technical Commission II, Hong Kong (Year).