2.1. Study Area and Sample Selection
Nanjing is the capital of the Jiangsu Province and a central city in eastern China. In 2020, the urban population of Nanjing was 8.09 million and its urbanisation rate was 86.8%, with a built-up area of 868.28 km
2. The main part of Nanjing covers an area of 278 km
2, which accommodates more than half of its urban population. According to 2019 statistical data, the average population density in the main city exceeded 15,000 people/km
2 (about 6000 dwelling/hm
2), with that in the central district exceeding 40,000 people/km
2. We chose Nanjing for the following reasons: (1) it is a typical densely populated area, meaning that our research results are generalisable to similar areas either in China or internationally, and (2) it has hundreds of pocket parks (including side green pockets, mini-parks, and street corner parks), which offer abundant samples for this study. For example, the development of pocket parks between 2013 and 2017 increased the per capita park area of Nanjing by approximately 10% [
39]. According to the newly formulated Nanjing Parkland Layout Plan (2017–2035), the main city will continue to promote the development of pocket parks, ensuring that every resident will eventually be able to access at least one park within a five-minute walking distance [
40].
The selection of pocket park samples was based on the following two criteria: (1) through considering the threshold area of the average community park in China (10,000 m
2) [
41] and the areas that have enough space for the flexible use of a pocket park [
42], the size of the pocket park sample was set as 800–10,000 m
2, and (2) to effectively collect the residents’ daily use information of the pocket parks, the pocket park samples must be located in the areas with residents in their surrounding 300 m (a five-minute walking distance) [
43]. Therefore, we obtained 74 pocket park samples, with an average area of 4102 m
2 and an average density of 31,000 residents/km
2 in the surrounding 300 m area (
Figure 1).
2.2. Data Collection
We collected the usage data of pocket parks through fixed-point shooting via a field survey. The investigations were carried out on weekdays and weekends (excluding national holidays) in both autumn (between late September and late November 2020) and spring (between late March and late May 2021). The climate of these periods in Nanjing is considered the most suitable for outdoor recreation [
44]. To collect the most representative usage data of pocket parks, we conducted the field surveys only on sunny or cloudy days, with the average temperature being between 15 °C and 25 °C. To trace the different usage rates of pocket parks over a given day, four periods, including early morning (6:30–8:30), late morning (9:30–11:30), afternoon (14:30–16:30), and night (18:30–20:30), were used as the target investigation periods because the use of pocket parks in these periods was found to be relatively more active. In contrast, the utilisation mode of pocket parks was relatively stable within each period, according to our preliminary site survey that involved 15 samples.
To improve the efficiency of our data collection, we divided the investigation area into five sub-areas according to the spatial distribution of the pocket park samples. Around 15 samples were located within each sub-area, which are relatively closed to each other and are convenient to be investigated as a group. One to two team members were responsible for data collection in each area, ensuring that the use conditions of each sample were recorded through photos in each target period during each investigation day. As existing studies have demonstrated that the factors affecting park use on weekdays and weekends differ [
33,
45], we collected the usage data of all the samples over 10 days, including five weekdays and five weekends.
The type of use and the number of users in each target period of every sample pocket park were recorded via photos taken during each field survey (
Figure 2). We classified pocket park use based on the six-category framework for outdoor activities in green spaces, as developed by Lachowycz and Jones [
46]. Three adjustments to the original framework were made to fit our sampled pocket parks and study objectives (
Table 1). Because the use category ‘interaction with wildlife and nature’ in the original framework was rarely found in our pocket park samples, we removed it. Second, we did not utilise the ‘travel through parks’ category as one of our flexible use categories for assessing pocket parks because it is not recreation-related. Instead, we counted people travelling through pocket parks as a variable termed ‘passer-by density’, which reflects the characteristics of the site conditions that may affect the flexible use of pocket parks. For example, a continuous linear space traversed within a pocket park can easily be used as a shortcut by pedestrians that could then influence the normal use of the given park. Third, according to the results of our site surveys, we expanded the category ‘other activities’ to include all activities that were not covered in the original categories, including standing and waiting, people being on their phone, and so on. We also found that different types of recreational activities were dominated by different age-groups. For example, most people participating in group activities are senior people (older than 60 years old), while most people involved in family and social interactions are children (younger than 10 years old) and middle-aged (30–60 years old). Therefore, the diversity of the recreation activities within a pocket park can also partially reflect the compatibility of a park to different age-groups. As a result, five categories of park-related uses, including physical activities, relaxation and entertainment activities, family and social interactions, participation in group activities, and other activities, were used to measure the use flexibility of pocket parks. In contrast, the original category ‘travel through parks’ was dubbed ‘passer-by density’ for measuring the site condition.
2.3. Model
We established two model series to analyse the influential factors and their effects on pocket park use. Each model series comprised two linear regression models, including a one-weekday model (Model 1), which was used to analyse the factors affecting pocket park use on weekdays, and a one-weekend model (Model 2), which was used to analyse the factors affecting pocket park use on the weekend. Model series 1 takes the indicator that reflects use flexibility as the dependent variable, which is used to analyse the factor types and their action effects on pocket park use flexibility. Model series 2 takes the indicator that reflects use intensity as the dependent variable, which is employed to analyse the factor types and their action effects on pocket park use intensity. Model series 2 also functions as a comparison to further outline the correlations and differences between the factors affecting pocket park use flexibility and those affecting use intensity (
Figure 3).
The use intensity of a pocket park in a specific period is measured by the visit frequency of the users counted from the photos taken in this period. To quantitatively measure the use flexibility of pocket parks, we introduced the Shannon Wiener index (SWI)—an index that was originally designed to evaluate the level of biodiversity in the field of ecology—to reflect the extent of the flexible use of parks by quantifying the diversity of their recreational activities in a specific period. The diversity measurement of the SWI uses the uncertainty measurement principle from information theory to predict the variations in the number of species that an individual may encounter within a specific community. The higher the diversity of the community, the greater is the variation in their natural species. The SWI of pocket park j (SWI
j) is derived from Equation (1).
where n is the total number of pocket park use categories and G
i is the proportion of the use intensity gained from each category I in terms of their total use intensity. The application of the SWI in the measurement of pocket parks’ use flexibility reflects the number of use categories and the distribution evenness of the users in each category. Accordingly, a higher SWI indicates higher use flexibility of each pocket park.
To be consistent with the attributes of pocket parks, we used the categories of park internal conditions and park surrounding conditions to aggregate the independent variables. The independent variables in each category were selected based on existing research [
27,
47] and then narrowed down using exploratory and correlation analyses. During this process, variables were either aggregated into a group or eliminated. For example, two preliminary variables in the surrounding category (the capacity of outdoor fitness equipment for adults and the capacity of outdoor fitness equipment for children) were aggregated into one variable titled ‘exercising facility capacity’, owing to their high collinearity. Meanwhile, for variables with high collinearity that could not be aggregated, the more influenced variable with a higher coefficient was retained. For example, the park area variable was found to be highly collinear with that of the paved ground area. After comparison, the latter was retained because it was found to be more influenced, while still reflecting more delicate features of the site condition. After screening, we extracted seven variables for modelling, including four internal condition variables and three surrounding condition variables (
Table 2 and
Table 3).
In the internal conditions category, paved ground usually accommodates most pocket park visitors, as reported in existing research [
34,
35]. To differentiate between pathways or passage spaces within parks, we included only paved ground that is larger than 12 × 12 m
2 in our statistics because these parameters are considered the minimum threshold size capable of supporting most group activities [
34,
42,
48]. The tree canopy coverage area was used because the space under a large tree has proven to be more attractive to park users because it improves the microclimate, as well as biodiversity [
49,
50]. To unify the measurements of various exercising facilities of differing sizes, we adopted a capacity unit to measure the exercising facilities (including fitness equipment for adults and children). As mentioned, passer-by density was also included as a dynamic attribute of the site conditions because it indirectly reflects the spatial structure within pocket parks. In addition to passer-by density, which was counted through a field survey, we measured the site condition data through the Maxar Worldview-03 images of Nanjing (April 2020), with a ground sampling distance of 0.51 m. Moreover, a site investigation was performed in October 2020 to measure the capacity of exercising facilities and to verify the borders of the paved ground, which cannot be observed directly from the Maxar images.
The surrounding conditions category included the boundary building ratio, resident population, and the land mixed-use level within the surrounding 300 m area of the pocket park samples. Approximately 300 m around each pocket park was defined as the surrounding area because this is considered the best service scope of pocket parks in most planning guidelines [
43,
51]. The boundary building ratio was included because it is an important indicator that reflects the spatial features of a pocket park (such as spatial closeness and openness to the city public space), as well as the potential complementary support that a pocket park can obtain from the boundary facilities, as per the existing guidelines [
52]/studies [
53]. The boundary building ratio was calculated through the land use maps and Maxar images. To measure the surrounding resident population, we captured the number of households in the surrounding area of each sample from the open-access website Lianjia, the largest letting and estate agency in China. Then, the resident population was calculated using the average household population of Nanjing, as derived from the statistical data of 2020 from the Nanjing Statistical Bureau. Land mixed-use level indicates the degree of land used in varying ways which is measured by land use entropy (LUE). LUE considers the relative percentage of two or more land-use types within a given area. When calculating the LUE, seven types of point of interest (POI) data, including shops, restaurants, houses, recreation and entertainment facilities, administrations and offices, medical agencies, and educational agencies, were also collected from the Baidu Map through the JavaScript Application Programming Interface. The POI data from the Baidu Map include the name, location coordinates, and subcategory of each facility type, thereby supporting the targeted geographic information system (GIS) mapping. Furthermore, these data are dynamically updated by app users, meaning that they are sufficiently accurate and timely to support our measurements. The LUE of pocket park j is derived from Equation (2).
where P
i is the percentage of the POI type i in the surrounding POI of pocket park j, with k being the total types of POI around pocket park j (by definition, k ≥ 2). A higher LUE indicates a greater extent of land mixed-use.
Moreover, to further investigate the usage modes of pocket parks in the samples, k-means clustering was used to segregate pocket park samples into two three-group series, according to the use flexibility (UF) and use intensity (UI). We used the relative mean value (clustering centre) of each group within the use intensity group and use flexibility group series to label each group as either ‘high value (H)’, ‘medium value (M)’, or ‘low value (L)’. Then, we segregated all the pocket park samples into nine usage modes, according to the combined results of the three labels, including the modes of H-UF and H-UI, H-UF and M-UI, H-UF and L-UI, M-UF and H-UI, M-UF and M-UI, M-UF and L-UI, L-UF and H-UI modes, L-UF and M-UI modes, and L-UF and L-UI. By analysing the distribution of samples in these nine usage modes and the user structure of each mode, we can reveal the service features of the pocket parks in the study area.