Urban parks are specifically intended and designed to offer diverse recreational service, which are particularly important for enhancing the health and well-being of urban residents [1
]. The use of the recreational services provided by urban parks largely depends on the movement of visitors from their homes to the parks [3
]. Thus, the accurate description of park visitors’ travel flows could help us to understand the actual spatial and temporal connections between parks and their beneficiaries. However, with the rapid development of cities, quantifying actual, specific travel flows to urban parks has become an increasingly complex task and particularly challenging in the face of unprecedented transportation networks, variations in individuals’ travel choices and preferences for particular experiences [4
]. Expanding knowledge of the travel flows of residents visiting urban parks is important for recreation planners and managers wishing to improve the efficient use of urban parks.
Previous research has quantified several spatial characteristics of the travel flows of urban residents using data from surveys and questionnaires [4
]. For example, Roorda and Ruiz [6
] analyzed the number and nature of travel flows by mode of transport, based on data from the Toronto travel activity panel survey in Canada, while Thøgersen [7
] reported travel route dependent behaviors with respect to choice of public transport using data from a Danish panel survey. Some studies have analyzed travelers’ attitudes towards alternative transportation, in order to provide information for urban planning and management [8
]. In addition, temporal variations in travel flows have been examined in recent studies, including differences in the behavior of residents between weekdays and weekends [10
] as well as seasonal variations [11
]. The majority of these studies typically focus on commuting, while travel for recreational purpose has been largely ignored, although lack of knowledge of these dynamics hinders understanding the interactions between urban parks and residents. Recently, spatial characteristics of tourist flows in a city, in particular international visitor flows, have been quantified in several case studies [12
]. Despite all this, the travel patterns of city dwellers visiting urban parks has still rarely been examined. Furthermore, limited attention has been paid to the overall characteristics of travel flows, such as flow paths, velocities, directions, times, and quantities. There is a need to examine travel flows of residents visiting urban parks in greater detail to understand the spatial and temporal connections between parks and residents.
With recent developments in mobile location technology, it is possible to track and record more information on individual travel flows. New technologies to obtain spatial-temporal travel data, such as GPS tracking [14
] and smart card recording technology [15
], have been applied in recent studies of travel behavior. For example, Smallwood et al. [16
] explored visitor movement patterns using travel networks in a large marine park in north-western Australia, while Orellana et al. [17
] revealed movement patterns of visitors in the Dwingelderveld National Park in the Netherlands using GPS technology. These new technologies have proven to be more accurate in collecting spatial and temporal travel data compared to surveys. However, most of the data collected using these new technologies cannot be directly linked to personal motivations. It is difficult to differentiate the motivation of recreational service use from any other travel purposes. Past studies have used various algorithms to compute scenic routes in order to identify recreational service flows within scenic spots, for example using Path Attribution Networks (SPANs) [18
], a single objective shortest path algorithm [20
] or road segment-based clustering of geo-tagged photos and Flickr images [21
]. Despite these studies having the clear purpose of examining recreational service use, and recreational service flows within scenic spots being clearly quantified, the spatial and temporal connections between scenic spots and residences still remain unclear. Moreover, not all visitors use these new technologies, and the representativeness of the results of these studies is limited. There is a need to combine traditional surveys and new technology in order to identify, comprehensively, the spatial and temporal characteristics of travel flows to urban parks.
Due to substantial spatial and temporal overlaps of individual travel flows, there is a view that travel convenience, time and experience constraints restrict the set of spatial-temporal opportunities for travelling to urban parks [23
]. By examining spatial and temporal bundles of travel flows, where different types of flow are co-located in the same route or occur at the same time, we can more accurately manage, both spatially and temporally, the synergies of multiple travel flows of residents visiting urban parks. Previous studies have identified spatial bundles of ecosystem services using correlation coefficient analysis in many cities [24
], while temporal bundles of ecosystem services are seldom identified. Studies on travel behavior have identified distinct travel groups using cluster analysis based on individual travel paths or socio-economic and demographic characteristics to represent homogenous groups [25
]. However, as travel flows of residents visiting urban parks have received limited attention, their bundles, therefore, have rarely studied, either in space or time.
Although spatial-temporal characteristics of activity-based travel flows have been quantified, few studies have attempted to focus on the characteristics of travel flows of residents visiting urban parks. This study seeks to fill this research gap by mapping and quantifying spatial and temporal dynamics and bundles of travel flows in Wuhan, P.R. China. The study was based on empirical investigation of individuals’ travel movements from their homes to the parks. The research was guided by the following two research questions: How are travel flows of residents visiting urban parks spatially and temporally distributed in the study area, and what groups of bundled travel flows can emerge in relation to space and time constraints?
2.1. Data Collection
2.1.1. Park Sampling
This study was conducted in the center of Wuhan, P.R. China. At the end of 2016, 33 free public parks were unevenly distributed across seven urban districts within an area of 863 km2
in the central part of the city. To exclude dilapidated and rarely used parks, we first selected a sample of 21 main public parks that receive more than 100,000 visitors per year, according to data from the Wuhan Municipal Bureau of Landscape and Forestry. We divided these 21 parks into three size classes: small (<10 ha), medium (10–50 ha), and large parks (>50 ha). We then randomly selected one park from each size category in each urban district, to ensure that our analysis covered different categories of park size and all the city’s central areas. Finally, a total of 12 public parks were selected, representing 57% of the 21 main public parks (including 67% of the small parks, 56% of the medium parks, and 50% of the large parks) in the study area. Figure 1
shows their locations.
To understand the travel behavior of people from their homes to the parks, face-to-face questionnaire surveys were conducted in the 12 selected parks during the summers of 2015 and 2016. Visitors passing research assistants or resting in places that were not too busy at the entrance inside the park were invited, at random, to participate in the investigation. However, only local visitors between 14 and 75 years old were included in the survey. The on-site interviews started with an introduction and clarification of the purpose of the study, and written informed consent was obtained from each participant before the questionnaire survey. We used a one-page questionnaire, which was discussed with an expert panel (the authors and five professors in relevant disciplines) in an effort to ensure it had sufficient face validity. A pre-investigation involving 50 respondents was conducted in Zhongshan Park to test the clarity of the questionnaire prior to conducting the formal survey. The final questionnaire was divided into two main sections. The first covered how visitors travel to the parks, including questions on their reasons for travel and sequence of movements, where they live, which park entrance they used, what times they arrive and leave the park, which mode of transport they use, and how often they visit. To comply with privacy policies, respondents were asked to write the location of the nearest bus stop to their residence instead of their home address. The second section of the questionnaire was designed to elicit information on respondents’ demographic characteristics, including their age, gender and occupation.
2.2. Mapping and Quantifying Spatial Travel Flows
The daily park travel flows of individuals reflect the spatial connections between urban parks and residents. Individual travel choices and movements are widely recognized to have significant effect on their travel flow patterns [17
], including travel flow paths, distances, velocities, quantities, and frequencies. Flow path (in this context) refers to the travel trajectory, from the location of an individual’s residence to the park entrance they used, while flow distance is the length of the flow path. The movement of people can be characterized by their travel velocity and frequency, which are determined by their choice of mode of transport and frequency of park visits, respectively. Flow quantity refers to the number of residents travelling to parks, and often declines with increasing travel distance. These dimensions are integrated together to form individual descriptions of travel flows.
The process of data selection was conducted before the mapping analysis. Only travel episodes directly from home to the park were considered. To estimate each respondent’s flow path, the shortest path along the road network connecting the park entrances they used to the bus stop closest to their residence was extracted using Baidu Map (a popular online search engine that Chinese urban residents use extensively for daily navigation), together with their reported mode of transport. Then, all extracted paths were converted into line shape file format in a geographical information system (GIS), and mode of transport, frequency, park arrival and departure time attributes were assigned to each of these lines. As described in most previous studies, mode of transport and frequency determine the inherent characteristics of travel flows [13
]. Mode of transport choice reflects different movement velocities and affects the experience of the trip for park use. Travel frequency is considered in terms of travel for recreational purposes, indicating individuals’ preferences. To further distinguish travel flows with inherent differences, we classified the flows into nine permutations of low, medium and high frequencies and velocities, thus separating the flows into groups that shared similar characteristics, as shown in the matrix presented in Table 1
. According to the questionnaire survey, visitors have four direct transport options for reaching urban parks: walking, driving, catching a bus, or taking the metro. We classified walking as the low velocity mode, and taking the metro as the high velocity mode, while driving and catching a bus were classified as the medium velocity mode, based on their relative speeds. According to common perceptions of residents in Wuhan, visiting a park more than three times per week (>150 times/year) was classified as high frequency, and visiting less than once a week (<50 times/year) was considered low frequency, with intermediate values classified as medium frequency. This classification system was selected because it is consistent with a common understanding and is easy both to communicate and understand. Next, nine corresponding sets of line shape files were extracted from the converted path files, mentioned above, using ArcGIS software.
To map spatial flows of travel to parks, a density surface of each flow path type was generated by kernel density estimation (KDE) analysis using the line shape file in ArcGIS. The KDE method has been widely used to describe spatial distributions of densities (of diverse variables) with distance by smoothing data converted into continuous density surfaces [26
]. In KDE analysis, a kernel’s bandwidth determines the accuracy of results [27
]. In total, six bandwidths were tested with the same output cell size (10 m × 10 m) and same kernel function in this study: 20, 50, 100, 200, 300, and 400 m. The inflexion point of the relationship between bandwidth and the maximum KDE value was used to estimate the optimal bandwidth for KDE analysis, and local hotspots of mapped events (here, flows) were displayed for verification. An output cell size of 10 m × 10 m was used because it provided sufficient precision for distinguishing individuals’ travel behavior from closed travel routes according to the spacing between adjacent roads in the study area. Finally, a kernel bandwidth of 100 m was employed, selected on the basis of the tests mentioned above. A flow chart of the procedure is shown in Figure 2
Descriptive statistics were used to characterize different types of travel flows, including length of flow path, travel time and frequency of park visits. For each variable, both the mean and standard deviation (SD) were calculated using SPSS software.
2.3. Identifying Temporal Dynamics of Travel Flows
To identify daily patterns of movement, we counted numbers of the respondents arriving and leaving in the questionnaire in one-hour intervals separately, based on the questionnaire survey. Scatter plots were then constructed to display the distributions of numbers of park visitors, and optimal fitting models (Peak, Gaussian, three parameters) were used to describe the temporal patterns for each category of travel flow. To illustrate the differences in patterns between arrival and departure times, the scatter plots and regression lines of the times were mapped together in the same figure for each category of flow. All these analyses were conducted using Sigmaplot 12.0 software.
2.4. Bundling Spatial and Temporal Travel Flows
As different types of travel flow may have similar spatial and temporal distributions, the coexistence of different types of flow with respect to space and time must be examined. This is achieved by applying the concept of “bundles” of travel flows that share similar characteristics. To explore spatial bundles of flows, we first converted road network lines generated by all travel paths into location points at 100 m intervals. Then, KDE values for each of the nine types of travel flow were extracted from the point shape files, thus nine KDE values were separately assigned to the attributes of each location point. These spatial location points were then used as samples for hierarchical cluster analysis to identify spatial bundles of travel flows, using the nine types of KDE values for each point as the cluster field. We ran agglomerative hierarchical cluster analysis using Euclidian distance and standardization of values to Z scores in SPSS 19.0 software. The hierarchical tree generated by this unification process was plotted to present a visual representation of the bundled groups of spatial flows.
Cluster and Outlier Analysis was used to estimate temporal bundles of travel flow paths that shared similar service times, using the Anselin Local Moran I statistic with inverse distance weighting in the ArcGIS environment. This procedure generates statistically significant clusters of high values (High-High), clusters of low values (Low-Low), outliers with high values surrounded by low values (High-Low), and outliers with low values surrounded by high values (Low-High). Here, lines of the flow paths were input as the analysis objects, while arrival times were used as the analysis field. Thus, High-High clusters, Low-Low clusters, High-Low outliers and Low-High outliers can be respectively regarded as flow bundles with late arrival times, flow bundles of with early arrival times, isolated flow paths of with late arrival times, and isolated flow paths of with early arrival times.
Quantifying travel flows of residents visiting urban parks, with respect to space and time, is important for understanding the processes involved in, and factors affecting, the recreational use of urban parks. This paper presents a visual impression of the spatial delivery and temporal dynamics of park visitor flows in Wuhan in summer, by mapping the movements of residents from their homes to parks, based on empirical investigation. The flow maps could help efforts to characterize park service efficiency and identify social inequities in park access from a user’s standpoint. Analysis of the temporal dynamics of travel flows showed that distributions of park visits are significantly time constrained. Thus, temporal dynamics of park visits should be considered in urban park management, especially peak arrival and departure times, which may help in the development of more dynamic strategies to meet different residents’ park visiting needs. Moreover, the analysis of bundles of travel flows from spatial and temporal perspectives, suggests that transport conditions strongly influence bundling of the spatial flows, whilst bundling of temporal flows is significantly related to spatial factors. These findings may help efforts to identify potential trade-offs and synergies among different types of travel flow, and facilitate exploration of policy alternatives to provide equitable and efficient urban park systems.