Spatiotemporal Patterns of the Use of Green Space by White-Collar Workers in Chinese Cities: A Study in Shenzhen

White-collar workers, with tremendous work pressure, excessive working hours, and poor physical condition, need green spaces not only to have physical exercise and social gatherings, but also to become closer to nature and to relieve stress for their mental health. In China, the 996 office schedule, working from 09:00 to 17:00 six days a week, has become popular in the workplace; under such high-intensity work and pressure, white-collar workers have limited time to access green space for leisure, and their use of green space for health benefits is compromised. This study selected Shenzhen Futian Central Business District to find out the green space use patterns and preferences of white-collar workers based on GPS data and questionnaire surveys. In addition, the value of green exposure in the time dimension was calculated according to individuals’ actual behaviors. Based on cluster analysis, this study summarized the typical green space use patterns of three groups of white-collar workers, which reflects the hidden inequity of white-collar groups who are subjected to varying degrees of spatiotemporal constraints in using green space. This paper puts forward three directions for the optimization of green space allocation, functional facilities, and improved walkability in employment-intensive urban areas. The results provide certain guiding significance for alleviating the mismatch of time and space in green space enjoyment and for improving the spatiotemporal inclusiveness of green spaces in urban central business districts.


Background
In the past decades, as the importance of accessibility to urban green space to public health has been recognized [1], uneven accessibility has become an environmental justice issue [2,3]. Given the correlation between green space and well-being, sufficient provision of and access to green space are considered two critical aspects of proper and healthy living conditions [1]. However, with the acceleration of urbanization, urban problems such as environmental pollution, traffic congestion, housing shortages, and a continuous decline in the quality of the urban ecological environment have surfaced. Urban green space is not fairly allocated to the residents who need it most.
Increasingly, more people are facing the prospect of living in an environment with fewer green resources, especially those in low socioeconomic groups, mainly including lowand middle-income or unemployed groups [2,[4][5][6], women [7,8], the elderly [9,10], and children [11]. At the same time, the distribution of public green space in most cities and towns is uneven. Influenced by CBD locations, the ethnicity of residents, and their relative wealth, education, and other attributes [12], provisioning of green space is disproportionate cities with a population of more than 1 million is 39 min a day [38]. The commuting problem has been exacerbated by huge increases in housing prices and continued migration from rural to urban areas [39]. Most employees who work by a 996 schedule live far from their workplaces, which means that they have to endure not only excessive working hours but also long commuting time [36]. Furthermore, there are new types of social stratification and differentiation among white-collar groups with the urban socio-spatial polarization in China [40]. Some white-collar groups have become marginalized due to relatively insufficient income, job stability, and social security. This also means further differentiation of vulnerable urban groups, which leads to a broader economic gap between different groups in the white-collar class, with significant differences in lifestyle and social identity [40][41][42].
The aim of this research was to investigate the spatiotemporal patterns of urban green space use in Chinese cities, in order to inform landscape planning and design in central business districts to benefit white-collar groups. With white-collar groups in Shenzhen as the research subjects, this study investigated the spatiotemporal patterns of urban green space use by examining the actual green activity paths of these workers in an employmentintensive area to evaluate the fairness of urban park supply and the distribution of facilities and services. Furthermore, given the growing economic gap due to the differentiation of the white-collar class in Chinese cities, a cluster analysis combined with socioeconomic attributes was conducted to depict green space use by white-collar groups of different classes and disclose the hidden inequity of groups who are subjected to varying degrees of spatiotemporal constraints in using green space.

Shenzhen's White-Collar Workers
Shenzhen, the first special economic zone (SEZ) in China, is known as a pioneer city of economic reforms [43]. Located in Guangdong Province in southern China, with an area of about 1990 km 2 and a population of 18 million, Shenzhen is the country's largest metropolis, with the highest population and employment density. Shenzhen's rapid economic growth has attracted massive numbers of young white-collar workers to live and work here. Meanwhile, due to the demographic polarization of vulnerable workers and wealthy elites, social-geographic isolation has become a prominent urban problem [44,45]. There are a number of social inequity issues, which are considered to be typical urban problems in developing countries [46,47]. The skyrocketing housing prices and growing job instability in Shenzhen, as a first-tier city in China, with many high-end financial and IT companies, force young workers to comply with the 996 corporate culture, which has a highly negative effect on their physical and mental health. Moreover, all city-level parks in Shenzhen are located in the SEZ, where the green space is positively correlated with housing prices [48]. Young white-collar workers can only live in distant areas, because they cannot afford the high housing prices in central urban areas, which is also proof of the inequitable use of green space. Despite their high education levels, these groups are struggling to find suitable jobs and achieve a standard of living that meets their expectations, rather than enjoying the supposed "middle class" lifestyle based on high income in the traditional sense [49]. The case of Shenzhen has high research significance for studying the green space use behavior of white-collar groups, which reflects the living conditions of special groups in China's megacities and provides a new research idea for exploring the environmental injustices in Chinese cities.

Shenzhen's Green Space Planning
The shortage of land is a huge challenge in Shenzhen, which has hindered the supply of green space in general [47]. The Chinese government has adopted a systematic form of urban planning and a series of regulations and policies to increase urban green space in the past two decades, and has made remarkable achievements in the supply of urban green space in Shenzhen and nationwide [50,51]. For example, since 2006, the Shenzhen government has established a three-level park construction system of "forest (country) Land 2021, 10, 1006 4 of 25 park-urban park-community park" and accomplished the goal of "a city of a thousand gardens" in 2020. However, planners and managers still have a limited understanding of social equality in the supply of urban green space. According to the research findings of You [47], areas in Shenzhen with a relatively concentrated mobile population (migrants) are more likely to have a short supply of green space, and socially and economically disadvantaged areas are subjected to more restrictions on access to green space. This is consistent with the research results of Tan et al. [52] on the distribution of public green space in Shenzhen at the sub-district level. Sub-districts with larger populations of lower socioeconomic status have fewer green space resources of inferior quality and with lower accessibility, presenting a disparity between the quantity and quality of green space [52].
Regarding the research on employment-intensive areas, no complete system or specific indicator requirements have been established in Shenzhen or across the country, with little research on the green space landscape supply in office areas. For years now, the greening rate, green coverage, and per capita park area have been used as the primary basis for evaluating urban greening in China's planning management systems [53]. They can be used to evaluate the total amount of urban green space intuitively and quantitatively [54]. However, there are significant differences among the green space indicators. Some scholars have interpreted the existing open space allocation specifications and suggested that there is a gap in the per capita allocation of public open spaces in employment-intensive areas [55]. Previous studies mostly used official datasets on land use or aerial photographs to measure the accessibility of public green space [56,57], or used per capita public green space as a quantitative indicator to measure the per capita public green space supply [56,58]. However, neither of them can measure the spatial relationship between a specific work-intensive area and the surrounding green space, let alone evaluate, guide, and control the distribution of greenery in work-intensive areas and the relationship between green spaces and office areas, especially regarding the problem of insufficient green space services.
At the same time, most quantitative studies on urban green space focus on residential and street green space and the quality of park green space at the static level [59,60], ignoring the tracking of actual exposure to green space in terms of user activity [61]. However, it is not enough to study environmental exposure only from the perspective of static space, as differences in individual activities often determine the different environmental impacts on individuals [62]. Therefore, how to accurately quantify the level of exposure of white-collar workers to green spaces in the natural environment based on their dynamic behavior trajectory has become a new problem in the study of green space in employment-intensive urban areas.
According to the concept of environmental exposure science, some scholars have proposed using the amount of green exposure as a new way to measure the effect of green space [63]. The method of simulating urban street scene images focuses more on evaluating green spaces by tracking the user's individual behavior from a human perspective [64], thus providing a precise way to calculate green exposure. However, at present, the research objects of this method are mostly urban residents, and the research areas are mostly community green spaces, urban parks, and urban street greenery. There is a lack of targeted discussion on the green exposure amount of special groups and the use of green exposure amount to guide the improvement of green space systems for special groups. Therefore, combining the behavioral characteristics of white-collar workers and the current situation of the layout of green space systems in employment-intensive areas, the study of green exposure can be based on the needs of white-collar workers and optimizing green space renewal strategies in these areas.

Study Area: Futian Central Business Distric
Futian is one of the central urban areas and the location of the Central Business District (CBD) in Shenzhen. As of 2019, Futian had a permanent population of 1.6337 million, ranking third in regional GDP among the top 100 urban areas in China. With extremely tight land resources, the park area in Futian District occupied up to 878.93 hm 2 , accounting for over 10% of the total area under jurisdiction; the green coverage area was 3384.48 hm 2 , with a green coverage rate exceeding 40%; and the total number of parks in the district was 119 [65].
As a typical case in China's first-tier cities, the Shenzhen Futian CBD was selected as the study case based on pre-investigation and a literature analysis (Figure 1). Based on the documents published by the Planning and Natural Resources Bureau of Shenzhen Municipality, Futian CBD refers to the area enclosed by Binhe Avenue, Lianhua Road, Caitian Road, and Xinzhou Road, with a total land area of 618.80 hectares. The green park space (G1) in the area is 230.47 hectares, the square area (G4) is 6.22 hectares, and the per capita area of community parks is 28.8-32.9 m 2 [66]. As the pilot CBD, merging the administrative center with the high-level cultural institution, Futian CBD has inspired the development of CBDs in Guangzhou, Hangzhou, Suzhou, Zhengzhou, and other cities. Its practices and experience have also profoundly influenced subsequent CBD construction in China [67]. For more than two decades, the Shenzhen Municipal Government has made every effort to achieve the goal of establishing a brand-new urban center that symbolizes the strategic ambition to build "a modern international city, a regional economic core city, a garden city, and the only financial, business, information, culture, and administrative center of the city" [68]. In summary, it is one of the most iconic areas representing the urban spatial image of Shenzhen, featuring "three highs": high-density population, highintensity development, and high-level construction [65]. It is also a typical urban office area, where many white-collar workers live and extensive green spaces are distributed ( Figure 2). Hence, it has certain research value and provides sufficient sample data for this study. Table 1 describes the status of the green spaces in the selected area.

Study Area: Futian Central Business Distric
Futian is one of the central urban areas and the location of the Central Business District (CBD) in Shenzhen. As of 2019, Futian had a permanent population of 1.6337 million, ranking third in regional GDP among the top 100 urban areas in China. With extremely tight land resources, the park area in Futian District occupied up to 878.93 hm 2 , accounting for over 10% of the total area under jurisdiction; the green coverage area was 3384.48 hm 2 , with a green coverage rate exceeding 40%; and the total number of parks in the district was 119 [65].
As a typical case in China's first-tier cities, the Shenzhen Futian CBD was selected as the study case based on pre-investigation and a literature analysis ( Figure 1). Based on the documents published by the Planning and Natural Resources Bureau of Shenzhen Municipality, Futian CBD refers to the area enclosed by Binhe Avenue, Lianhua Road, Caitian Road, and Xinzhou Road, with a total land area of 618.80 hectares. The green park space (G1) in the area is 230.47 hectares, the square area (G4) is 6.22 hectares, and the per capita area of community parks is 28.8-32.9 m 2 [66]. As the pilot CBD, merging the administrative center with the high-level cultural institution, Futian CBD has inspired the development of CBDs in Guangzhou, Hangzhou, Suzhou, Zhengzhou, and other cities. Its practices and experience have also profoundly influenced subsequent CBD construction in China [67]. For more than two decades, the Shenzhen Municipal Government has made every effort to achieve the goal of establishing a brand-new urban center that symbolizes the strategic ambition to build "a modern international city, a regional economic core city, a garden city, and the only financial, business, information, culture, and administrative center of the city" [68]. In summary, it is one of the most iconic areas representing the urban spatial image of Shenzhen, featuring "three highs": high-density population, highintensity development, and high-level construction [65]. It is also a typical urban office area, where many white-collar workers live and extensive green spaces are distributed ( Figure 2). Hence, it has certain research value and provides sufficient sample data for this study. Table 1 describes the status of the green spaces in the selected area.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of whitecollar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of whitecollar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of whitecollar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of whitecollar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of whitecollar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Data Collection and Processing
Modern technology can facilitate decision making with immediate and efficient ways to understand human spatial behavior [69]. In this study, the combination of a geographic information system (GIS) and the Global Positioning System (GPS) helps to monitor and record green space use behavior and analyze the coupling relationship between use path and green space distribution.
In this study, the green space use behavior and population attribute data of whitecollar workers in the selected area were collected by a combination of handheld GPS and questionnaires on working days from October 20 to November 10, 2020. The research team invited white-collar volunteers in the Futian Central District to participate. The sampling strategy was to find businesses in the area and invite white-collar workers in different occupations (technology, production, administration, marketing, R&D) and jobs (ordinary staff; grass-roots, middle, and senior managers), with different salary levels, and stratify their work and leisure time. Two research assistants were trained to collect the data. They recruited white-collar workers from office buildings in different parts of the study area for a one-day survey. The volunteers were asked to carry a handheld GPS device from 08:00 to 22:00 (about 14 h) in order to record their personal green space use behavior throughout the day. The GPS tracker was set to record time and space data, including latitude and longitude, every 20 s. After GPS data processing and screening, a total of 203 complete all-day travel trajectories were extracted.
Since this study is focused on the use of green space around the workplace by whitecollar workers, GPS data processing only refers their activities in the central area of Futian, and information on green space around their residences was not included in this study. According to the collected GPS data, from 8:00 a.m. to 10:00 p.m., all volunteers finished their use of green space in the Futian Central District, so the GPS tracking data in this period was retained. In order to further analyze the spatial and temporal rules of white-collar workers' use of green space around the workplace, the time was divided into five periods: morning (08:00-11:00), noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00).
The sample was concentrated in the 20-to-30-year-old age group, which shows that most white-collar workers in China are young. When the GPS trackers were collected the next day, a questionnaire survey was administered to collect personal socioeconomic information matching the trajectory, including age, gender, income, occupation, and educational background ( Table 2). The questionnaire also included the volunteers' leisure hours and the frequency of green space use around the workplace in the recent period (Table 3), as well as their subjective evaluation of the green space park proposal in Futian Central District. The GPS data and the survey were complementary to verify the validity of the use pattern.

Subjective Preference
Green spaces that are not adaptable to the preferences of various user groups will have a lower utilization rate, which is particularly evident in green spaces with fewer facilities and smaller area [70]. User preferences could help determine successful green space Land 2021, 10, 1006 9 of 25 design and affect how long people stay. Studies have confirmed that groups with different social and economic characteristics have different preferences and behavior patterns when using green space due to different motivations [8,13,[71][72][73]. Similarly, their socioeconomic background and the function and density of the area surrounding the park will also affect people's activities and use levels in the space [74]. This means that analyzing use behavior and understanding users' needs and preferences can help planners ensure the good quality of urban green space and improve its utilization rate. Based on a summary of many domestic and foreign articles, combined with the actual situation of parks and green spaces in Futian District, Shenzhen, the influencing factors of subjective evaluation are summarized in Table 4, including 22 factors in 5 categories (green space supply, traffic connection, facility functions, sanitation and management, and design factors). The questionnaire the volunteers filled out when they returned the GPS also included a rating of these factors on a scale of 1-5, with one being "strongly disagree" and five being "strongly agree". Table 4. Specific strategies of subjective evaluation.

Item Optimization strategy References
Green space supply

Analysis of Users' Green Exposure
As more attention is paid to individuals, scholars have begun to realize that there is a great deficiency when studying environmental exposure only from the perspective of static space [59,60]. Differences in individual activities often determine differences in the degree of environmental impact on individuals [62]. This study aims to fill this research gap and propose a research framework to measure green exposure by superimposing individual space-time movement tracks and the built environment. Based on behavioral tracking of white-collar populations in cities, urban street view images and manual photographs are used to quantify the amount of green exposure of the white-collar population on workdays.
First of all, the Futian Central District street view photo database needs to be established. In this study, five graduate students majoring in landscape architecture were recruited to take photos of street scenes in the area. In addition to the main traffic roads, all accessible roads, such as branch roads and lawn paths, were sampled. In order to avoid overlapping and cover the whole region, intersections of roads were also used as sampling points, and the distance between sampling points was moderate. In principle, sampling points were taken at intervals of 100 m. Important nodes in the iconic green spaces within the research scope were selected for manual photography.
Meanwhile, the collected photos were corrected and supplemented by Baidu Street View Map (https://map.baidu.com (accessed on 10 November 2020) to complete the database of street views and park photos in actual contact with the samples. Street view images were collected from four directions at each sampling point: Sight_Frontn, Sight_Rightn, Sight_Leftn, and Sight_Backn ( Figure 3). Each sampling point was numbered to facilitate subsequent calculation of green exposure based on sample GPS tracking. Given the difference in light during the day and night, photos were collected during both daytime and nighttime in this study. In the actual calculation, day and night photos at the same point were imported based on the actual times individuals were at certain points to calculate more realistic individual green exposure. This photo collection method, combining big data street view maps and manual photography, covered the whole research area to the maximum extent, making up for the defects of slow update speed and incomplete coverage of street view images, so as to more truly reflect the real-time built environment characteristics individuals are exposed to when they pass by.

Analysis of Users' Green Exposure
As more attention is paid to individuals, scholars have begun to realize that there is a great deficiency when studying environmental exposure only from the perspective of static space [59,60]. Differences in individual activities often determine differences in the degree of environmental impact on individuals [62]. This study aims to fill this research gap and propose a research framework to measure green exposure by superimposing individual space-time movement tracks and the built environment. Based on behavioral tracking of white-collar populations in cities, urban street view images and manual photographs are used to quantify the amount of green exposure of the white-collar population on workdays.
First of all, the Futian Central District street view photo database needs to be established. In this study, five graduate students majoring in landscape architecture were recruited to take photos of street scenes in the area. In addition to the main traffic roads, all accessible roads, such as branch roads and lawn paths, were sampled. In order to avoid overlapping and cover the whole region, intersections of roads were also used as sampling points, and the distance between sampling points was moderate. In principle, sampling points were taken at intervals of 100 m. Important nodes in the iconic green spaces within the research scope were selected for manual photography.
Meanwhile, the collected photos were corrected and supplemented by Baidu Street View Map (https://map.baidu.com (accessed on 10 November, 2020) to complete the database of street views and park photos in actual contact with the samples. Street view images were collected from four directions at each sampling point: Sight_Frontn, Sight_Rightn, Sight_Leftn, and Sight_Backn ( Figure 3). Each sampling point was numbered to facilitate subsequent calculation of green exposure based on sample GPS tracking. Given the difference in light during the day and night, photos were collected during both daytime and nighttime in this study. In the actual calculation, day and night photos at the same point were imported based on the actual times individuals were at certain points to calculate more realistic individual green exposure. This photo collection method, combining big data street view maps and manual photography, covered the whole research area to the maximum extent, making up for the defects of slow update speed and incomplete coverage of street view images, so as to more truly reflect the real-time built environment characteristics individuals are exposed to when they pass by.  After the image data were sampled, the green exposure in the behavior path was analyzed. An algorithm was developed in this study to calculate the green exposure of individual activity paths in multi-sampling images. In the analysis process, the image of a sampling point was first converted from RGB to HSV mode, and then the green image was extracted from the HSV spectrum (0-360 • ). The HSV street view image data that met the color requirements of the white-collar behavior trajectory were extracted by computer through this method. Figure 4 shows the whole process, including the original street view image (Figure 4a), manual extraction (Figure 4b), MATLAB software optimization extraction (Figure 4c), and optimization extraction after denoising through top hat operation (Figure 4d). Compared with manual extraction, the computer algorithm used in this study achieved a correct rate of 94.13%. Figure 5 shows the comparative effect of pixel extraction. After denoising, the image area obtained is the green area conforming to the exposure calculation.
After the image data were sampled, the green exposure in the behavior path was analyzed. An algorithm was developed in this study to calculate the green exposure of individual activity paths in multi-sampling images. In the analysis process, the image of a sampling point was first converted from RGB to HSV mode, and then the green image was extracted from the HSV spectrum (0-360°). The HSV street view image data that met the color requirements of the white-collar behavior trajectory were extracted by computer through this method. Figure 4 shows the whole process, including the original street view image (Figure 4a), manual extraction (Figure 4b), MATLAB software optimization extraction (Figure 4c), and optimization extraction after denoising through top hat operation (Figure 4d). Compared with manual extraction, the computer algorithm used in this study achieved a correct rate of 94.13%. Figure 5 shows the comparative effect of pixel extraction. After denoising, the image area obtained is the green area conforming to the exposure calculation.  Comparison effect of green pixel extraction algorithm. The purple part is the correctly extracted region of pixels, the red part is the actual green region that was not extracted, and the blue part is the non-green region that was incorrectly extracted by the algorithm.
Finally, according to the collected image information, the amount of green exposure of white-collar workers' according to their behavior tracks was determined, and the photos of sampling points of roads or park green spaces they passed by with the GPS were imported into the algorithm, with four photos for each sampling point. The time that the After the image data were sampled, the green exposure in the behavior path was analyzed. An algorithm was developed in this study to calculate the green exposure of individual activity paths in multi-sampling images. In the analysis process, the image of a sampling point was first converted from RGB to HSV mode, and then the green image was extracted from the HSV spectrum (0-360°). The HSV street view image data that met the color requirements of the white-collar behavior trajectory were extracted by computer through this method. Figure 4 shows the whole process, including the original street view image (Figure 4a), manual extraction (Figure 4b), MATLAB software optimization extraction (Figure 4c), and optimization extraction after denoising through top hat operation (Figure 4d). Compared with manual extraction, the computer algorithm used in this study achieved a correct rate of 94.13%. Figure 5 shows the comparative effect of pixel extraction. After denoising, the image area obtained is the green area conforming to the exposure calculation.  Comparison effect of green pixel extraction algorithm. The purple part is the correctly extracted region of pixels, the red part is the actual green region that was not extracted, and the blue part is the non-green region that was incorrectly extracted by the algorithm.
Finally, according to the collected image information, the amount of green exposure of white-collar workers' according to their behavior tracks was determined, and the photos of sampling points of roads or park green spaces they passed by with the GPS were imported into the algorithm, with four photos for each sampling point. The time that the Figure 5. Comparison effect of green pixel extraction algorithm. The purple part is the correctly extracted region of pixels, the red part is the actual green region that was not extracted, and the blue part is the non-green region that was incorrectly extracted by the algorithm.
Finally, according to the collected image information, the amount of green exposure of white-collar workers' according to their behavior tracks was determined, and the photos of sampling points of roads or park green spaces they passed by with the GPS were imported into the algorithm, with four photos for each sampling point. The time that the individual stayed at each sampling point was input. The calculation methods and steps of the algorithm are as follows: Step 1: Calculate the green exposure of a single image.
Land 2021, 10, 1006 12 of 25 Step 2: Calculate the green exposure at the coordinate point.
GreenLevel(EL n , NL n ) = Image_GreenLevel Sight_front (EL n , NL n ) × Weight f ront + Image_GreenLevel Sight_left (EL n , NL n ) × Weight le f t + Image_GreenLevel Sight_right (EL n , NL n ) × Weight right + Image_GreenLevel Sight_back (EL n , NL n ) × Weight back (2) where Weight f ront , Weight le f t , Weight right , and Weight back are the weighting coefficients in the four directions of front, left, right, and back during an individual's traveling process.
Step 3: Calculate the green exposure within the user's behavior path.
Considering the influence of exposure time in the same built environment on the green exposure value, the duration of the same coordinate point is weighted according to the GPS data collected during the calculation of green exposure. The measurement of green exposure can be more accurate when considering the time dimension. In an individual movement trajectory, the green exposure of all sampling points was multiplied and accumulated by the time spent at each point, and divided by the total travel time, to calculate one day of green exposure on the user's behavior trajectory. Each sampling point (coordinate point) consisted of four images, the time of which must be consistent in the software calculation.
where G n represents the amount of green exposure at a sampling point; t n represents the dwell time at a coordinate point, in minutes; n represents the number of coordinate points included in the calculation of the trajectory. Limited by the availability of data, the green visual exposure defined in this paper only includes individuals' activity routes and exposure to green parks, and does not include their exposure to green space when staying indoors (such as residence, workplace, etc.) or when traveling beyond the Futian Central Area.

Analytic Techniques
In this study, the variables were first divided into three categories based on the collected data to accurately classify the individual attributes and green space use characteristics of white-collar groups: subjective evaluation, green space use, and individual characteristic variables, representing subjective cognitive evaluation, actual use of green space, and individual natural and socioeconomic attributes, respectively. Through cluster analysis, the portraits of different green space users were verified by the classification model to summarize the typical behavior patterns of green space use by various whitecollar groups.
First, the k-means model was used for cluster analysis with SPSS software. The optimal clustering model was obtained by trying different numbers of clusters and comparing the results. Given that the number of samples was 203, the samples were grouped into clusters, and the number of clusters was controlled at fewer than 10. Modeling analysis was performed by setting 3, 5, and 10 clusters; by comparing the model results of these clusters, it was preliminarily shown that the 3-cluster model had significant differences between different variables in various clusters, suggesting that it was the most appropriate number of clusters (Table 5). Based on the classification of three clusters, a classification model was further established to verify its rationality and effectiveness. The sample ratios of the training set and the test set were set to 70% and 30%, respectively. Given the number of samples collected and the number of variables involved, the parent node was set to 30, the child node was set to 10, and the default CHAID was applied for the algorithm. Based on the three-cluster decision tree model obtained in this study, the variables used to distinguish types in the parent node included green space supply, time spent in the green space, and job title, representing subjective evaluation, green space usage, and individual characteristic indicators, respectively. This suggests that the variables designed and selected in this study were effective for the construction of the clustering and classification model. These variables were the primary basis for classifying different groups. Combined with the results shown in the prediction error and accuracy rate tables, the error rates of the training and test sets were 0.23 and 0.324 and the overall prediction accuracy rates were 77% and 67.6%, respectively. This suggests that the model results were relatively good and met the requirements for further analysis.
Meanwhile, the means of green exposure of the samples were compared based on the three clusters (Table 6). Furthermore, the GPS tracking data after screening were visualized on the STpath plug-in of the GIS9.3 platform to obtain a spatiotemporal path map of whitecollar volunteers enjoying green space around the workplace in Futian Central District on a certain working day (08:00-22:00). The visualizations used by visitors to the park help to generate hypotheses during the exploration phase and guide further statistical and spatial analysis to determine the optimal design, planning, and management strategies [92].

Results
After the rationality and effectiveness of the cluster analysis in this study were verified, an analysis was conducted based on the final cluster centers (Table 5) combined with the subjective evaluation of the three clusters extracted ( Figure 6) and the value of green exposure (Table 6) to explore the characteristics of green space use in employment-intensive areas and the subjective preferences of white-collar workers in the three clusters.

Results
After the rationality and effectiveness of the cluster analysis in this study were verified, an analysis was conducted based on the final cluster centers (Table 5) combined with the subjective evaluation of the three clusters extracted ( Figure 6) and the value of green exposure (Table 6) to explore the characteristics of green space use in employment-intensive areas and the subjective preferences of white-collar workers in the three clusters.

Portrait of White-Collar Groups Using Different Green Spaces
The first cluster comprised oldest overall age and highest education level and income in terms of individual characteristics. Given the sample characteristics of the area where the data were collected, this group can be summarized as middle and senior managers or successful people who are busy at work. In addition, the sample size of this group was only 48, accounting for the smallest proportion among all samples, which is in line with the positioning of elites. This type of population was in the middle of the three clusters in terms of green space use and reached the maximum mean in terms of average use time of green space per day. According to the data obtained from questionnaire statistics, 50% of the interviewees had an average total use time of green space per day of 30 min to 1 h. The reason may be that this type of user has high status in their work unit or team, so their working time is relatively flexible, with more free time. Their purpose for using green space may be related to their work, aside from resting and relaxing. They have sufficient

Portrait of White-Collar Groups Using Different Green Spaces
The first cluster comprised oldest overall age and highest education level and income in terms of individual characteristics. Given the sample characteristics of the area where the data were collected, this group can be summarized as middle and senior managers or successful people who are busy at work. In addition, the sample size of this group was only 48, accounting for the smallest proportion among all samples, which is in line with the positioning of elites. This type of population was in the middle of the three clusters in terms of green space use and reached the maximum mean in terms of average use time of green space per day. According to the data obtained from questionnaire statistics, 50% of the interviewees had an average total use time of green space per day of 30 min to 1 h. The reason may be that this type of user has high status in their work unit or team, so their working time is relatively flexible, with more free time. Their purpose for using green space may be related to their work, aside from resting and relaxing. They have sufficient time and motivation to be in green spaces. For this reason, a considerable number of people in the first cluster had a more obvious preference for small pocket parks (such as commercial affiliated green spaces). However, such small parks have a lower greening rate compared with large comprehensive or community parks, resulting in a higher standard deviation of green exposure for this group and significant differences in the values of green exposure for samples in the same cluster. The mean of facility function was much higher compared to the other four subjective variables (3.596), and that of the design factor was the lowest (1.963). The scores of factors in the five aspects of subjective evaluation by this group were significantly different. In terms of facility function, except for the factor of increasing security facilities being significantly higher than the other items, this group placed more value on functional than practical facilities compared with the aesthetic value provided by green parks.
The second cluster comprised the youngest age, lowest education level, and much lower income than the other two clusters in terms of individual characteristics. They also had the least rest every day, and all variables in the use of green space are at a low level. Those in line with these characteristics are fresh graduates who have just entered the workforce. Due to their low seniority and insufficient work experience and ability, their income is also at a low level, and their use of green space is limited by factors such as working hours or commuting distance. There were 80 samples in this group, indicating that they are the main group served by green spaces around the working environment. This group may have fewer exposure opportunities and shorter stays in green parks due to the constraint of insufficient daily rest time. Therefore, they have the lowest mean value for green exposure among the three clusters. There is little difference in the mean values of the five subjective evaluation indicators in this group. The mean of green space facilities is the highest (3.855), followed by green space supply (3.450), and that of traffic connection is the lowest (2.860). The analysis chart indicates that the mean of factors for improving security management in the venues was 3.95, which may be due to the insufficient sense of security in the use of outdoor green spaces among young people. This group's demand for venue security management also deserves attention.
In the third cluster, except for the average total use time of green space per day being slightly lower than that in the first cluster, the other variables are at the highest level of the three clusters in terms of green space use and subjective evaluation variables. There were 75 samples in this cluster, only five fewer than in the second cluster. Therefore, this cluster represents the main users of green space, and the most frequently used and highest rated. In terms of individual characteristics, it is at the middle level compared with the other two clusters and can be considered as the middle class of the samples and the most extensive and common intellectual or mental workers in this study, or the white-collar class in the traditional sense. Members of this cluster had higher incomes than fresh graduates, were more familiar with work, and had high work efficiency. Hence, they had more spare time at work, sufficient time and energy to use green space, and the highest use frequency and evaluation of green space. Compared with the first cluster, people in the third cluster have more choices for parks. In addition to commercial affiliated green spaces, they prefer comprehensive parks (central parks, community parks) with better green environments. They may also visit parks near their workplace at any time throughout the day. Therefore, their calculated mean green exposure is the highest among the three clusters. Among the five subjective evaluation variables, facility function is still the most important factor for this cluster (4.237), followed by green space supply (4.035), which is significantly higher than the other three factors.

Spatiotemporal Differentiation of Green Space Uses
Through classification, the spatiotemporal paths of green space use by the three clusters were extracted, and use during the day was divided into four time periods, noon (11:00-14:00), afternoon (14:00-17:00), evening (17:00-20:00), and night (20:00-22:00), to more clearly distinguish the paths of different white-collar groups and their time preferences for green space enjoyment (Figures 7 and 8). The vertical axis indicates the time of the day when the GPS data were collected, and the 2D plane is the edge map of the streets. The kernel density tool was used to obtain a kernel density heat map of the potential available green space distribution under the travel modes of three groups in terms of green space use (Figure 9). more clearly distinguish the paths of different white-collar groups and their time preferences for green space enjoyment (Figures 7 and 8). The vertical axis indicates the time of the day when the GPS data were collected, and the 2D plane is the edge map of the streets. The kernel density tool was used to obtain a kernel density heat map of the potential available green space distribution under the travel modes of three groups in terms of green space use (Figure 9).   more clearly distinguish the paths of different white-collar groups and their time preferences for green space enjoyment (Figures 7 and 8). The vertical axis indicates the time of the day when the GPS data were collected, and the 2D plane is the edge map of the streets. The kernel density tool was used to obtain a kernel density heat map of the potential available green space distribution under the travel modes of three groups in terms of green space use (Figure 9).    Figures 7 and 8 indicate that white-collar groups tend to use green space in the evening and night hours, and users have more choices in terms of green space types during these time periods and abundant activity paths. This suggests that in the evening and night hours, white-collar groups, free from fixed working hours, have more disposable time for rest, more freedom for activities, and more diverse purposes for green space use. Figures 7 and 8 indicate that white-collar groups tend to use green space in the evening and night hours, and users have more choices in terms of green space types during these time periods and abundant activity paths. This suggests that in the evening and night hours, white-collar groups, free from fixed working hours, have more disposable time for rest, more freedom for activities, and more diverse purposes for green space use.
The usage intensity of the central park and civic square in the evening and night hours was significantly higher than in the other periods. At noon and in the afternoon, the utilization rate of green parks was relatively low. Most interviewees conducted their activities in the vicinity of their workplace and were more inclined to choose green pocket spaces with commercial facilities; a small number of users may choose to take a rest in a park such as the sky garden on the top floor of Wongtee Plaza or the central park. The lowest rate of green space utilization during the noon period complies with the statistical analysis results of the sample data on average rest time per day, and 74% of the interviewees had only 1-2 h of rest during the workday. The results also indicate that most of the rest time on workdays was at noon, when most interviewees had lunch and lunch break activities. Since these activities can be completed inside the office building, few interviewees will go to a green space for relaxation. It is worth noting that the use rate at noon by the second cluster, which used green space less, was significantly higher compared to the other two clusters. According to the statistics, the number of people who used green space at noon accounted for 21% in the second cluster, 4% in the first cluster, and 8% in the third cluster. From Figure 9, it can be observed that the third cluster had higher use intensity of the civic square and the central park than the first and second clusters, which validates the results from the spatiotemporal path distribution map. The third cluster chose green parks more frequently, with a larger range of activities.

Connotation of the Inequity of Green Space Use among White-Collar Groups
The results of this study indicate that the spatial patterns of green spaces in Futian CBD enjoyed by various white-collar groups differ in terms of coverage range, use intensity, and free time. The classification and characterization of white-collar groups also suggest that groups whose use activity path covers a smaller distribution area of green space are characterized by a more significant time shortage in terms of green space enjoyment. This indicates that in addition to the subjective preference factors of users, spatiotemporal constraints have an important influence. Being subjected to the dual constraints of time The usage intensity of the central park and civic square in the evening and night hours was significantly higher than in the other periods. At noon and in the afternoon, the utilization rate of green parks was relatively low. Most interviewees conducted their activities in the vicinity of their workplace and were more inclined to choose green pocket spaces with commercial facilities; a small number of users may choose to take a rest in a park such as the sky garden on the top floor of Wongtee Plaza or the central park. The lowest rate of green space utilization during the noon period complies with the statistical analysis results of the sample data on average rest time per day, and 74% of the interviewees had only 1-2 h of rest during the workday. The results also indicate that most of the rest time on workdays was at noon, when most interviewees had lunch and lunch break activities. Since these activities can be completed inside the office building, few interviewees will go to a green space for relaxation. It is worth noting that the use rate at noon by the second cluster, which used green space less, was significantly higher compared to the other two clusters. According to the statistics, the number of people who used green space at noon accounted for 21% in the second cluster, 4% in the first cluster, and 8% in the third cluster. From Figure 9, it can be observed that the third cluster had higher use intensity of the civic square and the central park than the first and second clusters, which validates the results from the spatiotemporal path distribution map. The third cluster chose green parks more frequently, with a larger range of activities.

Connotation of the Inequity of Green Space Use among White-Collar Groups
The results of this study indicate that the spatial patterns of green spaces in Futian CBD enjoyed by various white-collar groups differ in terms of coverage range, use intensity, and free time. The classification and characterization of white-collar groups also suggest that groups whose use activity path covers a smaller distribution area of green space are characterized by a more significant time shortage in terms of green space enjoyment. This indicates that in addition to the subjective preference factors of users, spatiotemporal constraints have an important influence. Being subjected to the dual constraints of time and space, white-collar groups have become vulnerable in the macro-environment of the 996 overtime culture in Shenzhen with regard to green space enjoyment.
Previous studies mainly focused on specific dimensions, but did not comprehensively analyze possible inconsistencies in providing public green spaces and supplying facilities to meet the needs of different social groups. For example, the existing methods of evaluating urban green space mostly focus on assessing static objective data from the bird's eye view, and the subjects are mostly ordinary urban residents, while the special group of white-collar workers and the real green exposure in their dynamic activity trajectories are ignored. In this study, using green space as a breakthrough point, the actual level of green space enjoyment by white-collar workers is evaluated effectively through data mining of their dynamic activities of green space use. The differences in the relationship between the individual behavior paths of white-collar groups in the 996 overtime culture in Shenzhen and the distribution of green spaces in the central district are analyzed in depth mathematically to create a social portrait of special urban groups. The inequity in the provision of green space and the function of facilities used by white-collar groups under the dual constraints of time and space is verified, which indicates the potential conflict between the demand, supply, and utilization of urban green spaces in Shenzhen's employment-intensive areas. Urban planners should start with the current supply and use of public green spaces and consider whether the real demand can be met.

Optimization of Green Space Planning Based on Spatiotemporal Characteristics
According to the park usage density map, no matter what kind of clustering is used, Wongtee Plaza, located in the south axis of the central district of Futian, is the area with the highest utilization rate of green space by the white-collar population, and the all-day use intensity is very high. This is because the plaza is centrally located, surrounded by dense office buildings, which can provide relatively convenient transportation conditions. In addition, the green space in this area has certain commercial facilities, so there is highintensity use of the space between the shopping mall and office buildings. In addition to convenient food services and the rest and relaxation needs of white-collar workers, it can also provide opportunities for them to enjoy green space during busy working hours (noon and afternoon) and expand their activities such as interpersonal communication and work negotiation. Based on an understanding of the spatial characteristics of green space used by white-collar workers, the reasons for the high visit rates in this area were further analyzed with regard to their use purpose and subjective preference, so as to provide some reference for the optimization of green space allocation in employment-intensive areas. This paper describes the characteristics of green spaces preferred by white-collar workers and the possibility of future optimization from three aspects:

1.
Small-scale green space with pleasant landscape According to the spatiotemporal path analysis of users, due to the limitations of the workplace and time, white-collar workers' use of green space is scattered. Based on the motivation of these individuals to use green space, the purpose of their visit during working time is related to the nature of their work, which differs from the purpose of leisure during off-work time. Compared to large parks, pocket green spaces such as Wongtee Plaza distributed among dense buildings are more suitable for white-collar workers to conduct business negotiations and social interactions when their time is fragmented. At the same time, this study shows that increasing the number of large parks and green spaces does not promote their use by white-collar groups. In contrast, small pocket parks are more popular. The results provide direction for optimizing the design of green parks in employment centers. Due to the current long-term land use pattern, strong land demand, and other restrictions, it may be difficult to build large public green spaces in compact central areas [50]. Therefore, other greening strategies, such as vertical greening, parkways, and greenways, can be considered to create better green space and more opportunities for use [78]. For example, Singapore's Landscaping for Urban Spaces and High-Rises Programme (LUSH), implemented since 2009, creatively includes vertical green spaces in planned green areas. It expands the coverage of green space substitution regulations and represents an improvement in environmental quality in high-density urban areas [93].
Small pocket park afforestation should also improve at the same time, in view of the popularity of green space for commercial use by white-collar workers during the week. Accessory greenbelts located in shopping malls or commercial buildings around green spaces provide a better environment and facilities, but unavoidably, this kind of open space on rigid pavement is too much, causing a greenery scarcity problem. Some studies have found that if users are paying attention to the hard landscape, it will affect their perception of vegetation and greenness [94]. The type and proportion of hard landscape should be reasonably considered to blend its size, color, shape, and other attributes with the environment. Appropriately reducing the proportion of hard landscape can promote residents' use of urban green spaces [95]. Meanwhile, there is a positive correlation between the proportion of vegetation and residents' perception of greening [96,97]. Therefore, in order to increase white-collar workers' exposure to green spaces and improve their sense of the green experience, under the condition of not violating the building construction specifications, green plants and flower beds can be appropriately added, and green spaces with various types of vegetation can be reasonably configured to form differentiated features from hard spaces, so as to attract people's attention and encourage their use and enjoyment.

2.
Park configuration with diverse functions and complete facilities Studies have confirmed that inadequate facilities are among the important factors restricting the use of green space [87]. One reason for the popularity of the pocket green space in Wongtee Plaza is that the site has relatively complete infrastructure, including sufficient commercial facilities, which is crucial to attracting people to visit and rest [86]. The commercial facilities and green space in the park complement each other to some extent, jointly affecting the attraction of the site, providing temporary catering, social, and entertainment activities for white-collar groups. At the same time, the three groups in our study show that in the pursuit of green park "practicality", considering the characteristics of white-collar group green space use in the different periods, in addition to providing green land supply to alleviate the time constraints, perfecting green space facilities should be considered, especially comfortable and service facilities, to enhance the attractiveness and utilization of green space. On the one hand, for green spaces during working hours, improving the experience should focus on the low-intensity use at noon and in the afternoon. In addition to providing facilities for shade and cooling off in the sun-exposed park to improve people's experience, non-commercial tables and chairs and smoking areas could be added to green spaces and streets in work-intensive areas to improve the experience of people with limited time.
On the other hand, it is equally important to provide safety facilities for evening use. This study shows that evening is the peak time for white-collar workers, which is consistent with Ngesan and Karim's findings. Busy lifestyles and hot climate change the behavior patterns of urban communities, making people's leisure time shift to after sunset [98]. However, the current planning and design of urban parks are not entirely suitable for nighttime recreation in community parks. For example, some scholars found that in parks without police patrols, users will have a sense of insecurity, leading to a negative experience [83]. Through field investigation, this study found that the lack of lighting facilities in the central park and sky garden in the Futian Central District means there is dim lighting, which makes people, especially women, feel slightly insecure when doing leisure activities at night. In the calculation of green exposure among the sample in this study, it was also verified that in the same green space, if there is sufficient light, people's perception of green at night is lower than during the day. This also confirms that night lighting is important to green exposure not only because of safety issues but also for greater green perception. In addition, for the groups who seldom used green spaces during working hours, safety factors had higher mean values in the subjective evaluation. Therefore, in order to prolong use time at night, it is very important to enhance the spatial vitality of less popular green spaces, strengthen the security at night, and improve the quality of lighting.

Convenient and accessible transportation facilities
Wongtee Plaza is located in the center of the southern part of the central district. The road layout of small blocks and a dense road network makes the roads accessible and convenient for white-collar workers in the surrounding office buildings. In addition, the abundance of bus stops makes the plaza a rest station for workers between the workplace and their residence after work, which increases the opportunity to use the green space for those who are short on time. It can be seen that the accessibility and convenience of transportation around the green space can affect the utilization rate of white-collar workers. The central area of Futian provides sufficient green space with the north-south landscape axis layout. This space is well connected, but the central axis of the landscape has insufficient guiding power for white-collar workers in densely populated office buildings on the east and west sides. Considering that white-collar workers tend to use green spaces within walking distance on weekdays, in order to strengthen the connection between the office area on the east and west sides and the central landscape axis, it is more important to improve the traffic efficiency around the green space and reduce the negative impact of vehicle traffic on the workers' walking experience. An air corridor system is a good way to solve this. It can create a vibrant pedestrian environment, providing vantage points for pedestrians to enjoy different views of the city landscape, with its functions and paths enhancing the image of the city [99]. Especially in a high-density city, with an air corridor directly connected to offices and commercial and public open spaces forming a continuous walking space, users can avoid the inconvenience of crossing roads and can travel between buildings and parks, which greatly improves the degree of user mobility and convenience, which would be higher in the Futian Central District, where the feasibility of land use is limited.

Conclusions
In this study, through clustering model construction and classification model verification, different green space users were divided into three clusters based on the background of the study area and the individual characteristics of the samples. The first cluster included successful people with a high-level position at work and high requirements for green space. The second cluster included fresh graduates just entering the workplace who used green spaces less frequently. The third cluster included the main users of green spaces, with a certain amount of work experience. In terms of green space use variables, the order third cluster > first cluster > second cluster was observed in general. A comparison shows that the green exposure of the clusters also agrees with this feature. For the subjective evaluation variables, the order third cluster > second cluster > first cluster is observed. In the subjective evaluation by the three clusters, facility function is the most prominent factor, which is significantly higher in the evaluation of members of the first cluster than the other four aspects. In terms of the spatiotemporal characteristics of green space use, the distribution of potentially available green spaces changes according to the group that enjoys them and presents different spatial patterns: 1.
The use of green space by white-collar groups has evident spatiotemporal patterns, concentrated in the evening and night hours, when these groups enjoy a wider range of activities. Green parks are least used at noon, and mainly by the second cluster; in the afternoon, users have the smallest range of activities and the highest path concentration, limited to green spaces around their workplaces.

2.
The three clusters of users have insufficient use and uneven use intensity of green spaces distributed in the Futian CBD. The preferred parks are the commercial affiliated green spaces of the south-central axis. In particular, the surrounding green spaces with Wongtee Plaza as the center have the highest intensity of use. The scope of use is extended northward only in the evening and night hours.

3.
Even within the same cluster, due to individual differences in the time and preference of green space use, green exposure can vary significantly. Users who prefer commercially affiliated green space and are accustomed to using green spaces at night have significantly less actual green exposure than those who prefer comprehensive and community parks and who are accustomed to visiting parks during the daytime.
On this basis, this study proposes the following green space optimization strategies using the strategies summarized in Table 4. For the first cluster group, the measure with the highest mean value of subjective evaluation is "Provide facilities to improve the comfort of green spaces" (4.08), followed by "Add service facilities" (3.98), which is the same for the second cluster group, with the mean values of 4.18 and 4.06 respectively. Therefore, this paper puts forward suggestions to improve the configuration of green space park comfort and service facilities. Considering that the second cluster group of people have fewer opportunities to use green space, sunshade and rain protection facilities as well as water mist spraying facilities should be added to the green space near the workplace to improve the thermal comfort of using the park green space in hot weather (noon and afternoon). Although these items are also above middle level in the subjective evaluation of the third cluster group, the item with the highest mean value of subjective evaluation of the third cluster group is "Improve the sanitation of green spaces" (4.65), which is related to the management and maintenance of park green space and should also be paid attention to as an optimization suggestion by park managers. At the same time, combined with the preference of the second and third clusters on weekdays, the targeted planting strategy is feasible. This means that resources can be focused on specific target street greening. For example, in the densely populated area of office buildings, planting more street trees or increasing vegetation coverage can improve the visual greenness of time-short people during working hours, and provide good outdoor green space for workers to relax in the interval.
In general, as mentioned above, as the "vulnerable group" of green space use that has been neglected in the field of green space research, the white-collar group should be the main users of park green space in an urban high-density central area. The most important thing is to save the time cost of those who are short of green space and increase their access to green space. Therefore, it is urgent to increase the coupling of individual spatial and temporal path range and workplace green space distribution to alleviate the spatial and temporal mismatch of green space enjoyment. For example, further increase the small-scale green space between office buildings in the central district, including platform garden and roof garden, so as to achieve "green space in every gap". At the same time, enhancing safety management during the night time can make green space better fit into this group of people's time frame.
There are still some deficiencies in this study. First, although this paper uses street view and manual photos to establish a database and proposes a method of measuring green exposure based on individual mobility, the update speed of street view photos is slow and the coverage is limited to main urban roads, resulting in a large amount of work needed to supplement the database with manual photos. Therefore, compared with updated data acquisition methods (such as sensors, wearable cameras, etc.), the accuracy of measuring exposure to the built environment in terms of individual space-time movement tracking is lower. Therefore, advanced methods such as wearable cameras [100] and eye-tracking technology [101] will be considered in subsequent studies to obtain more detailed data of the built environment, so as to measure individuals' actual green environment exposure. Second, limited to the availability of data, this study chose only the Shenzhen open space office area as a typical case for discussion. Future research could include comparisons with other high-density office areas with white-collar workers using green spaces and dig deeper into the relationship between the variables, in order to determine whether the use of green spaces by white-collar groups is more universal, so as to promote the appropriate design and improve the utilization rate and satisfaction of urban green parks. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study. Data Availability Statement: Not applicable.