With the acceleration of urbanization in China, residents' demand for urban green space (UGS) is growing (1). Meanwhile, as an important component of urban ecosystems (2), UGS can reduce the negative impact of environmental factors such as space pollution (3), noise (4), and high temperatures (5). Prolonged exposure to UGS helps promote the physical and mental health of residents (6) and reduces stress as well as the risk of obesity, stroke, and other diseases (7). Due to that the traditional evaluation of UGS is based on the traditional indicators of urban park area per capita, the greening rate and green coverage rate, it can be difficult to reflect the spatial distribution, quality, and fairness of the service function of UGS (8). In the context of limited urban construction land, it is of great theoretical and practical significance to evaluate the spatial distribution quality of the existing UGS.

Accessibility is an important index to measure the fairness of UGS spatial distribution. Accessibility refers to the potential of overcoming resistance, such as time and distance, to reach a certain place (9). In recent years, an increasing number of studies have introduced the concept of accessibility into the evaluation of UGS. Researchers often measure the accessibility of UGS by setting statistical indicators (10) and using the cost-weighted distance method (11) or the gravity model method (12). The statistical index method evaluates accessibility by counting the number of places and area of UGS in the region, or by counting the number of places and area of UGS within a certain buffer zone from the living area (13). This method is also known as the cumulative opportunities method (14) or the covering method (15). The disadvantage is that it does not take into account the resistance that residents encounter on their way to UGS, and it is difficult to reflect the accessibility within the UGS' actual service area. Based on remote sensing data, the cost-weighted distance method assigns different crossing resistances to different urban landscape types, and the accessibility is evaluated by calculating the cumulative resistance encountered by residents on their way to the UGS (8). The advantage is that this method takes into account the impact of resistance on accessibility, but the disadvantage is that it does not consider the attractiveness of UGS of different qualities and levels; hence, the accessibility calculation results cannot accurately reflect the actual recreational situation (16). In the calculation of accessibility, the gravity model (17) adopts the distance decay function and considers the service capacity and potential of UGS, so that the results can better reflect the impact of UGS' attraction on accessibility. However, the relationship between residents' demand and accessibility remains to be further explored (18). The two-step floating catchment area (2SFCA) method was first proposed by Radke in 2000 (19) and was further improved by Luo and Qi (20) in 2009. Accessibility decreases with the increase of the resistance encountered by the residents on their way to the UGS and increases with the increase of the area and quality of the UGS. Based on the 2SFCA, the accessibility of UGS can be calculated comprehensively and simply from the perspective of the interaction between the park supply and residential demand, which makes up for the shortcomings of the gravity model (21). Due to its advantages, this method has been widely used in accessibility evaluation (21–24), and many studies have tried to improve this method to promote the accuracy of the evaluation results (Table 1).

Some studies revealed the competition effect between supply facilities and proposed the 3SFCA model to address the fact that 2SFCA cannot reflect the actual supply and demand situation. Based on the traditional 2SFCA model, 3SFCA calculates the option value between each pair of demand points and supply facility points and measures the competitive effect of multiple supply points within the same demand point search radius (25). However, Delamater showed that 3SFCA overestimates the competitive effect among supply points while ignoring the availability of the facility resources (26). Therefore, a simpler Modified 2SFCA model was proposed, which introduced the attenuation function in the selection weight to reflect whether facility resources are fully utilized, simplifying the quantification process of competition effect between supply facilities. Luo et al. (27) introduced the Huff model to reflect the selection weights, and on this basis, proposed the Huff 2SFCA model, which quantified the probability of people choosing supply facilities with continuous weight functions, further improving the applicability of the 2SFCA model. In the original form of 2SFCA, distance decay is treated dichotomically (28): the accessibility within the search radius is the same, and the accessibility outside the search radius is 0. This method ignores the influence of the spatial distance on accessibility. Some studies used different distance decay functions to solve this problem. For example, Luo and Qi (20) proposed the optimized 2SFCA to segment the distance within the search radius. However, the segmentation of distance is greatly influenced by subjectivity to a certain extent, and the segmentation weight given to the distance will lead to discontinuous results at the segmentation boundary. Therefore, 2SFCA applies continuous functions such as the gravity model (29), kernel density function (30), and Gaussian function (31) to solve the above problems. The Gaussian function is an s-shaped attenuation, and the attenuation rate of accessibility with distance is slower in the near and far stages than in the middle, which is more consistent with the actual situation than the 2SFCA method. Therefore, it is widely used in the optimization of accessibility evaluation methods (32). The 2SFCA model also fails to consider multiple traffic modes comprehensively. Yang et al. (23) proposed the multi-mode 2SFCA model to solve this problem. This model uses the weighted average time of multiple transportation modes to replace the time of a single transportation mode, which can better reflect the heterogeneity of accessibility under different transportation modes. Taking the central urban area of Hefei as an example, this study optimized the 2SFCA model for the purpose of comparing the accessibility differences of UGS under various transportation modes. Mobile phone signaling data were used in the study and combined with the Gaode map path planning model to improve the shortcomings of poor timeliness and low accuracy of the evaluation results caused by the use of statistical data for evaluation in traditional research. The UGS recreational situation extracted from the mobile phone signaling data and accessibility evaluation results was used to calculate the UGS supply and demand deviation to further elucidate the fairness of UGS spatial distribution and optimization strategies.

In this study, mobile phone signaling data were used instead of traditional statistical data to obtain accurate population positioning data. The distance between supply and demand points was represented by the actual distance of the Gaode Map path planning tool, which strengthened the data granularity and improved the accuracy of the evaluation results. The optimized Ga2SFCA method was used to measure the accessibility level under different transportation modes, and the actual traffic mode proportion data were used to calculate the comprehensive accessibility level. In addition, two different search distances for the time threshold were set based on residents' comfort and limited travel time. The accessibility differences under different travel modes and time thresholds were discussed. Finally, we combined the results of the UGS accessibility evaluation with the actual use of UGS to analyze the difference between supply and demand. In this study, we focused on the application of mobile phone signaling data in measuring the actual usage of UGS. Based on mobile signaling data, we acquired the precise resident positioning data in the residential area and the actual recreational activities in the UGS. Combined with the path planning interface in the Gaode API, with the support of refined travel data, we can analyze the coupling matching degree between the number of UGS visitors and the accessibility level results to explore the actual situation of the UGS supply and demand relationship. The use of fine-grained data and the improvement of the 2FSCA method are the most important contributions of this study.

In general, the grids with high accessibility to the UGSs in the central urban area of Hefei tend to be distributed near large urban parks, and this phenomenon is more obvious in the evaluation results of walking and cycling accessibility at the 15-min time threshold. However, with the increase in the time threshold and travel distance, the accessibility evaluation results gradually show a regular distributed pattern. In addition, the accessibility level of the driving mode at the 15-min time threshold and the walking and cycling mode at the 30-min time threshold shows that the number of grids above the moderate level increases significantly. For some socially vulnerable groups, especially low-income groups and the elderly, the ownership of a private car and the ability to undertake long-distance travel are clearly unfair conditions for access to more UGSs (42, 43). Therefore, it is still necessary to increase the area of UGSs to increase the opportunities for urban residents to obtain green space resources and ensure fairness of access to UGS for socially vulnerable groups. When urban land resources are limited and incremental expansion of old urban areas is difficult, decision-making departments should first consider improving the quality of existing UGSs. In densely populated areas with limited vacant land, smaller street gardens, pocket parks, and rooftop gardens should be considered to improve the accessibility of the UGS. Decision-makers should increase the number of large park entrances and strengthen the construction of roads around comprehensive parks.

In addition, China's government recommends that residents stick to low-carbon travel modes, which means giving priority to walking, cycling, and public transportation. For the old urban area, where there is limited urban construction land, it is undoubtedly a better choice to further develop the low-carbon transportation system than to increase the green area of the park. The construction of pedestrian and cycling networks, as well as the public transportation system, should be strengthened to improve the connectivity between residential areas and UGSs. According to the specific situation, decision-making departments should consider increasing the residential land area around UGSs to improve the utilization rate of UGSs in the new urban area, where the resources are sufficient.

The fairness of the UGS layout is the basic guarantee of urban residents' quality of life. Based on the optimized Ga2SFCA method, we evaluated the spatial accessibility of the existing UGS in the central urban area of Hefei under different travel modes and analyzed the supply and demand deviation of the UGS combined with the actual recreational population. The results show the following:

(1) Setting different travel modes and different time thresholds significantly affects the evaluation results of the UGS accessibility level. At the 15-min threshold, the areas with a moderate or higher walking comfort accessibility level are small, and the areas with a moderate or higher driving comfort accessibility level are far larger than those with walking and cycling comfort accessibility. At the 30-min threshold, there are relatively fewer areas with a moderate or higher level of walking comfort accessibility, while the areas with moderate or higher cycling accessibility were far greater than the former. The spatial distribution of UGSs in the central urban area of Hefei is uneven. In general, with the increase in the time threshold, the number of grids above the moderate accessibility level generally increases, and the spatial distribution of the grids with moderate, high and highest accessibility level present different patterns of contiguous, clusters, and spots distribution respectively. The accessibility of UGSs in old urban areas is poor, mainly because these areas have a relatively dense population and a large demand for UGS. Due to the shortage of construction land, it is difficult to add new UGS, even if the problem of the relatively small number of UGSs becomes more severe.

(2) The characteristics of supply-demand deviation of UGSs are revealed by analyzing the coupling matching degree between the number of UGS visitors and the accessibility level results. At the 15-min threshold, the demand of UGSs in the central urban area is greater than the supply, but at the 30-min threshold, the supply exceeds the demand. In addition, at the two different time thresholds, the number of grids with balanced supply and demand is small and the grids are scattered, mostly distributed around large comprehensive parks. Such areas usually have moderate residential density and good road connectivity. The grids with balanced supply and demand or more UGS supply than demand are mainly concentrated in the southern area and the edge of the central city near the suburbs. All these areas have large, comprehensive parks. The population density of the surrounding communities is moderate, and the road connectivity is better, with a higher level of transportation infrastructure than that in the old urban area. The southern area is the new urban area established by Hefei City, which has more capital and land for the construction of UGSs and various urban infrastructure, and the construction level of living environment is relatively high. At the same time, an increasing number of people choose to live in the new urban area, which further brings capital and population vitality and reduces the supply-demand deviation. Due to the increasing urban population, the population density of the old urban area is not decreased, which makes UGS resources scarce. The government should pay attention to improving the existing quality of UGSs, as well as alleviating UGS inequality by using small vacant lands and inefficiently developed lands to increase the area of UGSs.

Refined and scientific evaluation of UGS accessibility is an important foundation for the accurate restoration of urban green space fairness. Based on the refined mobile phone signaling data, we can obtain the precise accessibility evaluation results and actual supply and demand situation, which can certainly help the government to understand the status of the actual park usage. For the area with imbalanced supply and demand, the government should carry out thematic studies and further explore the reasons. This study can provide guidance for the government to formulate fair green space policies, advocate to guarantee access to green space resources for the socially vulnerable groups, and finally promote the healthy and sustainable development of cities. This study also had some limitations, largely due to the availability of data. When calculating the supply capacity of UGS, we did not consider the internal facilities, service quality, and service facilities of UGS. In the calculation of UGS demand, only the population number of the statistical unit was considered, without an in-depth consideration of the needs of people with different social attributes, which can be significant (44). A discussion on the causes of the discrepancy between the supply and demand of accessibility is needed and will be further explored in future research.

The data analyzed in this study is subject to the following licenses/restrictions: The urban green space data were derived from the Gaode API (application programming interface) (https://lbs.amap.com/), which is publicly available. The travel mode and travel distance data was also obtained through the path planning interface in the Gaode API. Additionally, the residential population and recreational visitor data were extracted from the China Unicom mobile phone signaling database (http://daas.smartsteps.com), which is not publicly available, and, after logging into the database system, the data can be retrieved and statistically analyzed by using structured query language. Requests to access these datasets should be directed to Gaode API (application programming interface) (https://lbs.amap.com/); China Unicom mobile phone signaling database (http://daas.smartsteps.com).

JC and CW contributed to the conception and design of the study. CW performed the statistical analysis and wrote the first draft of the manuscript. JC wrote sections of the manuscript. YZ and DL contributed to funding acquisition. All authors contributed to manuscript revision, read, and approved the submitted version.

This research received financial support from Anhui Agricultural University (Grant No. RC372016 and No. K2137001) and Anhui Provincial Department of Science and Technology (Grant No. 202004a06020014).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.