Urban park accessibility assessment using human mobility data: a systematic review

ABSTRACT

The rise in human mobility data in recent years provides a new frontier to understanding the spatiotemporal dynamics of mobility to urban parks. These human mobility data offer profound merit in the study of how accessible urban parks are through real-time analytics of individual travel patterns and locations. This study systematically reviewed 46 peer-reviewed publications from the Web of Science and Academic Search Complete databases relevant to human mobility data and urban park accessibility measures. The study aims to provide a comprehensive and quantitative evaluation of the utilization of human mobility data in the field of urban park accessibility research. This objective is pursued through a systematic examination of existing literature within this research domain. We summarize the types of mobility datasets and the modelling tasks adopted. Meanwhile, we grouped the methodological/thematic frameworks from empirical use of human mobility data in park accessibility into seven categories: 1) inequality and inequity level, 2) users’ perception and exposure level, 3) frequency and variations in park visitations, 4) park service area or effective service radius, 5) happiness and sentiment level, 6) travel mode choice and trip assignment, and 7) park characteristics. Additionally, we highlighted the obstacles in integrating human mobility data into park accessibility research and engaged in a discussion surrounding prominent ethical dilemmas related to the utilization of big data.

KEYWORDS:

• Urban park
• accessibility
• human mobility
• systematic review

1. Introduction

The scholarly literature on urban green spaces, including urban parks, has witnessed a remarkable surge in growth. Although both urban parks and urban green spaces are used interchangeably in many studies, they are notably different. Urban green spaces encompass all types of planned spaces within urban settings such as urban parks, neighbourhood parks, street and roadside greens, community gardens, cemeteries, and trails. Whereas urban parks on the other hand denote specific urban spaces devoted to recreation, relaxation, and leisure. Urban parks are distinguished by their design criteria such as park size, population threshold (i.e. the minimum capacity of people that urban parks are designed to accommodate during the planning and design stages), location, and facilities and amenities. The benefits of urban parks have been well documented to include; health (Brown, Schebella, and Weber 2014; Dzhambov and Dimitrova 2014; Kothencz et al. 2017; Shuvo et al. 2020; Stoia et al. 2022), recreation, physical activity, social interaction (Aziz, van den Bosch, and Nillson 2018; W. Liu et al. 2021), ecological services (Almeida et al. 2018), economic values (More, Stevens, and Allen 1988) and urban beautification (Sa’adu Danjaji et al. 2018).

The issue of urban park accessibility has garnered substantial attention within the scholarly community, resulting in numerous published studies. Many of these studies have developed accessibility frameworks based on specific indicators, such as the per-person index, exposure index, and minimum walking distance index, frequently utilizing traditional census data, satellite data, and survey sample data as data sources (Abdulraheeem et al. 2022; Konijnendijk 2021; World Health Organization 2017). However, the emergence of widespread mobile phone usage in the early 2000s ushered in a new era of data collection, commonly referred to as the digital data era. This transformative period has given rise to the concept of human mobility data where billions of individuals carry mobile phones with them at all times, engage in social media networking, post comments on e-commerce platforms, and utilize navigation applications – all of which generate vast quantities of data on human mobility (Zhao et al. 2016). While there is not a universally accepted definition of ‘human mobility data’, attempts to delineate its scope are widespread in the scholarly literature (Barbosa et al. 2018; Smolak et al. 2021; Zhao et al. 2016). It is common to find human mobility data being used interchangeably with the term ‘big data’, although it is important to note that human mobility constitutes a subset of big data. While all human mobility data can be considered big data, the reverse is not necessarily true (A. Wang et al. 2021). In this study, we operationally define human mobility data within the context of big data as large-scale datasets that capture the spatial movement and interactions of individuals within or across multiple geographic locations. This encompasses data acquired through a variety of sources, including but not limited to, mobile devices, GPS trackers, social media check-ins, and transportation networks, thereby enabling comprehensive insights into patterns, trends, and predictive models of human mobility on a granular level. The exponential growth in human mobility data presents a unique opportunity to examine various aspects of urban human mobility patterns. For instance, consider a park visitor who uses a rideshare service to reach a park; this individual enters details about the park’s location and specifies pickup and drop-off times. Alternatively, imagine a parkgoer who walks from home to the park and the phone’s location is actively on to record the movement at set intervals. Consequently, scenarios of this nature within a city generate an immense volume of spatiotemporal data, often reaching into the billions of data points.

Some recent studies have already started taking advantage of this big data era to investigate diverse urban subjects like traffic forecasting systems (Goh et al. 2012; Jung, Wang, and Stanley 2008), epidemic control (Belik, Geisel, and Brockmann 2011; Ni and Weng 2009), urban sustainability (Kong, Liu, and Wu 2020), and urban planning (Qi et al. 2011; Yuan, Zheng, and Xie 2012; Zheng et al. 2011) among others. These studies are powerful in that they provide useful insights that traditional research approaches could not have provided given the improving state-of-the-art machine learning models. Regarding the topic of urban park accessibility in the context of this paper, our focus is exclusively directed towards the utilization of tangible visitation records sourced from timestamps on social media, mobile phone GPS trackers, map navigation applications, and transactional records commonly referred to as digital footprints (although digital footprint is loosely used in this context). This is because urban park accessibility has transcended the subject of how many urban parks exist in an urban centre but instead how many persons utilize or have access to these parks as shown in the aggregate spatiotemporal data points generated. Whereupon in this study we categorize urban park use and park visit as forms of urban park accessibility.

Hence, the primary aim of this study is to offer a comprehensive and quantitative evaluation of the utilization of human mobility data in the field of urban park accessibility research. This objective is pursued through a systematic examination of existing literature within this research domain. Subsequently, we endeavour to address and elucidate three pivotal research inquiries:

1. What are the diverse types and scales of human mobility data employed in the analysis of urban park accessibility?

2. What methodological/thematic frameworks have been explored to evaluate urban park accessibility using human mobility data?

3. What challenges and limitations are entailed in the utilization of human mobility data for assessing urban park accessibility?

To address these research questions, our investigation entails an exhaustive review of 46 peer-reviewed publications, meticulously selected from a pool of 865 articles procured from the Web of Science and Academic Search Complete databases. These chosen publications specifically pertain to the integration of human mobility data in the assessment of urban park accessibility. Furthermore, we categorize the methodological/thematic frameworks into seven overlapping classifications and delve deeply into the obstacles and ethical issues with the adoption of human mobility data in research studies on park accessibility.

2. Theoretical underpinnings of accessibility

The concept of space accessibility is inherently multidimensional, encompassing the nuanced understanding and measurement of the relative ease with which individuals and goods can reach diverse locations in space. Within the existing body of literature, various models have been developed, and for the purposes of this study, we anchor our exploration within three spatial pipelines: 1) activity-based modelling, 2) social justice and equity, and 3) smart cities and technology. To establish a foundational understanding, accessibility, a term frequently employed in transportation planning, emerges as an abstract concept with debatable objective quantification (Handy and Niemeier 1997). Mathematical models are often constructed by researchers to evaluate their value, with a key emphasis on the interrelationship between land uses and transport systems. A notable definition characterizes accessibility as ‘the ease and convenience of access to spatially distributed opportunities with a choice of travel’ (US Department of the Environment 2005). However, quantifying the notions of ‘ease and convenience’ proves challenging due to geographical, economic, and social-equity considerations (Dong et al. 2006), a challenge resonating strongly in the context of park accessibility.

Activity-based models delve into individual daily activity patterns and travel behaviour, offering a utilitarian-based perspective on accessibility to parks (Nahmias-Biran et al. 2021). This model assesses the utility derived by individuals by factoring in the economic distance to a point of interest (e.g. park) from home or census-track level, however, the theory is limitedly applied. In contrast to traditional accessibility measures, which often rely on geographic proximity metrics such as minimum distance, service radius, and per capita indexes (Jin et al. 2023), we propose an enhanced model that incorporates both distance and utility derived from real-time travel data – a concept explored in Dong et al.‘s transportation policy research (2006), adapted for park-based accessibility.

The social justice and equity approach introduces an additional dimension to accessibility measurement by aggregating and disaggregating socioeconomic characteristics on a considerably scalable basis. Rather than relying solely on static registries such as census data and opportunistic surveys, this approach leverages granularity data from mobility platforms to conduct upscaled random analyses of park visitation in urban areas, considering the changing dynamics states of cities.

The third dimension, smart cities, and technology, represents a burgeoning model facilitated by advanced urban computing and the Internet of Things (IoT). This model integrates data, sensors, and digital infrastructure to enhance land use systems, particularly in transportation, and to track human mobility (Ang et al. 2022; Z. Li, Chen, and Yan 2021). In operationalizing the definition of accessibility to urban parks, we employ a framework that integrates human mobility data within these three dimensions, encapsulating park use and visitation. Our focus is on assessing the relative ease and convenience for all socioeconomic categories of the urban population to utilize or visit parks, gauging utility through real-time data. Our literature review extensively covers scholarship aligned with this multifaceted approach (Figure 1).

Figure 1. The conceptual diagram of this review.

3. Methodology

3.1. Article selection

In line with existing conventions of systematic review research, we adopted the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) model developed by Moher et al. originally for the health care journals which has become generally accepted in all scientific review reporting to identify, scan, and select relevant papers within the scope of our research (Moher et al. 2009). Abiding by previous works (e.g. P. Liu and Biljecki 2022; J. Wang and Biljecki 2022), we obtained papers relevant to human mobility data and urban park accessibility from Web of Science (WoS) and Academic Search Complete (EBSCOhost) databases reckoned for popularity within the scientific research community based on sets of defined search syntax discussed in Section 3.2.

3.2. Search criteria

The search criteria follow the set of defined syntax. For the paper topics, we prompted the advanced search engine in both WoS and EBSCOhost to search through paper topics, including titles, keywords, and abstracts relevant to human mobility data types and urban park accessibility. The Boolean search has three sets, i.e. first set: ‘urban park* OR public park* OR municipal park* OR green park* OR city park* OR local park*’, second set: ‘accessibility*, and third set: mobility* OR visit* OR travel*’. Note that the wildcard symbol ‘*’ helps identify the variations of the search terms. We restricted our review to peer-reviewed journal articles written exclusively in English. Concerning the disciplinary focus of the papers, we curated our search within the Web of Science (WoS) database, specifically targeting the following fields: environmental sciences, environmental studies, urban studies, urban and regional planning, sustainable greening, sociology, civil engineering, geography, and interdisciplinary domains. It is important to note that the EBSCOhost database does not afford the facility to encompass such a wide array of disciplines. We set our publication timespan to commence from 2010, primarily because the first commercially accessible Android smartphones equipped with internet connectivity and Global Positioning System features were introduced in 2008. The choice of this timespan also aligns with the rise in popularity of social media, artificial intelligence, big data analytics, and the Internet of Things (IoT), which have become increasingly influential in the given context since the start of the designated period.

3.3. Eligibility criteria

Our initial article pool resulted in 865 papers, which is reasonable considering the scope of our research and the growing focus on sustainability and sustainable cities. The papers that successfully progressed through the ‘Identification’ and ‘Screening’ stages were then evaluated based on specific eligibility criteria. Upon a thorough review of the 134 articles, we identified 12 duplicates. Additionally, 78 of the articles did not incorporate human mobility data within their methodological frameworks for assessing urban park accessibility, thereby failing to meet the eligibility criteria. Consequently, these articles were excluded from further consideration. This process resulted in a remaining corpus of 44 papers. Subsequent to this selection, an additional two papers were identified via snowball sampling, bringing the total count of evaluated manuscripts to 46. Despite the fact that a majority of the disqualified papers were pertinent to the topic of urban park accessibility, employing conventional methodologies such as self-administered survey questionnaires, census data, and other static data sources, they were excluded due to the lack of human mobility data as the focal point of analysis. This subset constituted 9% (equivalent to 78 papers) of the total manuscript pool examined in our study. Furthermore, our review identified a significant number of manuscripts that extended their focus beyond urban parks to encompass national parks, green spaces, regional parks, open areas, among others. To maintain the specificity of our research scope, we utilized the ‘not’ operator in our search criteria to effectively exclude these categories (suburban park” AND ‘urban parking’ AND ‘open space’).

In the end, our result yielded 46 peer-reviewed papers. We downloaded the paper list that contained comprehensive information about each of them including, title, abstract, authors name, year of publication, and journal. Our result of less than 6% (46/865) of the initial article pool is robust without bias-two main criteria saw many articles excluded; these are several duplicate records and the absence of human mobility data types as the main object of focus. We present the PRISMA workflow for our article selection in Figure 2.

Figure 2. PRISMA workflow for article selection.

4. Results

4.1. Descriptive statistics

The statistics of information for the papers scanned are provided as detailed statistical charts in Figure 3. These statistics of information include the publication year, spatial distribution of the publication, modelling tasks, and the different types of human mobility data.

Figure 3. Statistics of papers showing, (a) the number of papers published per year; (b) human mobility data types used in urban park accessibility; (c) spatial distribution of countries where the studies were conducted (study regions); (d) broad modeling tasks and methods used.

Figure 3(a) demonstrates a progressive increase in the utilization of human mobility data. This upward trend indicates the growing popularity of human mobility as an emerging domain field since 2010, particularly after notable studies conducted by Long et al. (2015), Vazquez-Prokopec et al. (2013), and Wu et al. (2016) in disciplines such as cultural geography, public health, transportation, and urban studies, including smart cities (A. Wang et al. 2021). Notably, in 2022, there were 16 publications (34.8% of the total) that employed human mobility data for the study of urban park accessibility. The rise in such publications can be attributed to COVID-19 in 2020. Scholars have extensively investigated the various impacts of COVID-19 control measures, encompassing both pharmaceutical and non-pharmaceutical interventions such as stay-at-home orders (X. Huang et al. 2022). Since 2020 to the present, researchers have focused on comprehending the effects of these measures on the dynamics of urban park visits prior to, during, and after the pandemic era, as explored by Geng et al. (2021), Jay et al. (2022), Linnell et al. (2022), Rice and Pan (2021), and Sung et al. (2022).

Figure 3(b) illustrates the various types of human mobility data currently employed to evaluate urban park accessibility. Four primary categories of human mobility data were used in prior studies, including social media, navigation services, mobile phone data, and transit records. Among these, social media emerges as the most prevalent platform utilized by researchers, accounting for 39.1% of the total reviewed publications. Mobile phone data and navigation services closely follow, with 14 and 13 publications, respectively. Conversely, transit records such as smart cards, transit cards, and go-cards exhibit limited representation, with only one publication devoted to their application.

Figure 3(c) presents the geographical distribution of research endeavours across seven distinct countries. The distribution of these studies is notably uneven, with a significant concentration observed in China, followed by the United States (U.S.). Out of the 46 reviewed papers, 27 are based in China, with a particular focus on densely populated cities such as Beijing, Chengdu, Shanghai, Wuhan, Nanjing, and Shenzhen. This geographic emphasis stems from the presence of historical parks and ring roads in Beijing (Qin et al. 2020), the abundance of flora and fauna in Chengdu (Tan et al. 2022), and the proximity of the Yangtze River to Wuhan, Nanjing, and Shanghai. In contrast, the utilization of human mobility data for urban park accessibility is less extensively documented in the United Kingdom (1 publication), Japan (2 publications), Singapore (2 publications), South Korea (1 publication), and a solitary global-scale record.

The modelling tasks depicted in Figure 3(d) provide insights into the methodological approaches employed by researchers to assess urban park accessibility from a human mobility data perspective. In this study, we comprehensively classify the modelling tasks into seven overlapping categories, which are extensively elaborated upon in Section 4.3. Notably, regression models stand out as the most frequently adopted modelling task in a significant portion of the examined studies, with a total of 15 publications. Next, clustering and spatiotemporal analysis methods emerge as the most employed tasks, with 9 and 8 studies respectively. Meanwhile, 4 studies amalgamate correlation and regression models within their methodological frameworks, while an equal number of studies adopt machine learning techniques. Specifically, they apply Natural Language Processing (NLP) to train and scrutinize user expressions pre and post-park visitation, a process often termed sentiment analysis. Intriguingly, only a singular study makes use of topic modelling.

The statistics are based on the 46 peer-reviewed articles that made the eligibility phase and were used for the study.

The statistics are based on the 46 peer-reviewed articles that made the eligibility phase and used for the study. In Figure 4(a), among microblogging social media platforms, Twitter emerges as the predominant human mobility data utilized in studies of urban park accessibility. This prevalence can be attributed to Twitter’s open policy (unfortunately the open policy has been modified following the purchase by Elon Musk in October 2022), which enables researchers to access users’ tweets through the official Twitter streaming Application Programming Interface (API), as demonstrated by Huang et al. (2021) who collected data from over 2.9 million unique Twitter users in the U.S. between 1 January 2020, and 30 June 2020, to evaluate residents’ travel trajectories during the COVID-19 pandemic. Numerous other studies have leveraged Twitter data as real-time human mobility data on various scientific urban studies, but we have only considered those relevant to our research (Chuang, Benita, and Tunçer 2022; C. Li et al. 2022; Schwartz et al. 2022). Next is the Chinese Sina Weibo social media platform, which accounts for 5 out of 18 social media data types. Sina Weibo, launched in 2009 by the Sina Corporation, has seen its user base grow to over 50 million (Yu, Asur, and Huberman 2011). This platform has become a valuable source of scientific mobility data, particularly in capturing users’ sentiments during the COVID-19 pandemic (S. Wang, Wang, and Liu 2021) and exploring health topics in China more generally (A. Wang et al. 2021). The remaining social media platforms – Instagram, Tencent, and Flickr – have provided data sources for 3 studies. The limited utilization of these sources can be ascribed to several factors within the academic context. Firstly, Instagram presents challenges for web crawling due to constraints on accessing users’ data. Secondly, extracting contextual information from Flickr is particularly demanding given the dominance of photos over textual content on the platform. Finally, there is a paucity of available case studies pertinent to Tencent User Data (TUD) that align with the specific scope of our research.

Figure 4. Statistics of papers that use (a) social media data, (b) navigation services data, and (c) mobile phone data.

In Figure 4(b), we present the different types of navigation services data employed. These data types primarily fall into three categories: Google Mobility Map, Chinese Map, and Baidu Map, which are discussed in Section 4.2. Baidu Map, a popular Chinese online navigation search platform, provides navigation information such as travel routes, mode choices, shortest route paths, and users’ travel history, like Google Maps. Remarkably, 64% of studies utilized Baidu Map through its API. On the other hand, the Google Mobility Report derived from Google Location History (and map) provides human mobility data and is associated with the largest online map service globally, catering to billions of users worldwide. However, only 27% of studies employed Google Mobility data. This discrepancy can be attributed to the spatial distribution of our research, with over 50% of publications originating from China, where Baidu and other data sources are predominantly used instead of the Google Mobility Report. Amap, another navigation service provider based in China, was utilized in a smaller number of studies, accounting for 9% of the publications. Although Amap is a leading digital content map provider, navigation, and location-based solutions in China, its usage remains relatively low compared to the other navigation services.

Figure 4(c) presents the various types of mobile phone data employed in studies focusing on urban park accessibility. These data types can be broadly categorized into three groups based on their source: SafeGraph data, data obtained directly from telecommunication companies, and mobile-installed applications (Apps). Notably, mobile phone data obtained directly from telecommunication companies emerged as the predominant data type among the three types, with 9 out of 14 studies. For instance, exemplary studies sourced data from the Chinese Mobile Telecommunication Company (Y. Zhai et al. 2018), the Korean Telecommunication Company SKT (Sung et al. 2022), Shanghai China Mobile Company (Xiao, Wang, and Fang 2019), and NTT DoCoMo in Japan (Guan et al. 2020). In contrast, anonymized mobile phone data from SafeGraph, as discussed in Section 4.2.1, was only employed in three case studies conducted in the U.S. This limited usage can be attributed to the fact that SafeGraph provides foot traffic data and POI data exclusively for the U.S. and Canada, while it offers only POI data for other countries. Thirdly, mobile-installed apps utilize built-in GPS software development kits (SKT) to collect locational mobility information from individuals using specific mobile phone applications, as observed in Linnell et al. (2022).

4.2. Human mobility data used in urban park study

Human mobility data as defined in the introduction section are mostly anonymized and aggregated. These datasets are generated through location-based services (LBS), which encompass a range of sources such as mobile phones, navigation maps, social media platforms, and sensor devices like smart cards and connected vehicles. A substantial portion of these datasets originates from mobile phone usage, including smartphone applications for social media and navigation services, telecommunications signals, and detailed call records. However, in the context of this study, we opt to categorize these data into distinct classes based on their methods of generation, forming subgroups, rather than grouping them solely under the parent class of ‘mobile phone data’. Consequently, even though social media and navigation services are applications embedded within mobile phones, we treat them as discrete categories. On that basis, we identify four primary sources of human mobility data frequently employed in the scholarly literature to evaluate urban park accessibility. These sources are as follows: 1) mobile phone data; 2) social media data; 3) navigation services data; and 4) public transit records.

4.2.1. Mobile phone data

Mobile cellular networks, commonly known as mobile phone signals or call detail records (CDR), are valuable sources of human mobility data. When individuals make calls or engage in browsing activities, cellular network providers record various details, including user ID, time, coordinates, and call duration. These details provide insights into the movements of mobile phone users over time and across different locations. Over the years, mobile cellular networks have evolved with improved network coverage, faster data speeds, extended communication options, and mobile IP configurations. This advancement enables service providers to collect mobility information of subscribers within the signal range of mobile base stations. Researchers have leveraged these records from cellular network providers to study human mobility patterns, including dwelling times in urban parks (Linnell et al. 2022). This approach is particularly efficient and cost-effective in countries with extensive network coverage, such as China and Japan, where major providers like China Mobile and DoCoMo serve large portions of the population (Guan et al. 2021; Y. Zhai et al. 2018).

To identify park users through cellular network data, researchers typically make assumptions about user behaviour. They filter out inactive mobile stations, focus on stations within or near parks, and define park activities based on recorded tower data. Some researchers also require that park visitors have recorded trajectories and spend a reasonable amount of time within the park (Guo, Song, et al. 2019; Ren and Guan 2022). In this study, three categories of mobile phone data sources were explored: SafeGraph data, data from telecommunication providers (discussed above), and data from mobile phone-installed apps. SafeGraph provides anonymized data on user origins, destinations, trajectories, and points of interest (POIs) for millions of mobile phone users, primarily in the U.S (Guan et al. 2021; Y. Song et al. 2022). SafeGraph collects location information through GPS and extracts data from internet users and government open sources. Data from mobile phone-installed apps is less commonly used due to its reliance on app-specific software that collects location information when mobile hotspots are active. Only a few studies have employed this approach, including one comparing home-based and activity-based exposure to urban parks in New York and another exploring park visitation during the COVID-19 pandemic (Linnell et al. 2022; Yoo and Roberts 2022).

4.2.2. Social media data

Social media data has become a valuable source of information for studying human mobility patterns, urban park accessibility, and related factors (X. Huang, Wang, Yang, et al. 2024). This data is generated through various social media platforms, including Facebook, WhatsApp, Twitter, LinkedIn, WeChat, Wiki, Instagram, TikTok, and YouTube. The transition from Web 1.0 to Web 2.0 has resulted in the proliferation of user-generated content (UGC) on these platforms, leading to the accumulation of massive amounts of user-generated data (X. Huang, Wang, Lu, et al. 2024).

One platform that has been extensively used for studying urban park accessibility is Twitter (now ‘X’). Schwartz et al. (2022) collected over 1.5 million tweets from 25 major U.S. cities to assess the happiness levels of users within and outside of parks. They used sentiment analysis algorithms to analyse the emotional content of tweets, finding that park users generally expressed more positive sentiments compared to non-park users. Twitter’s live streaming API allows for the collection of geographically tagged user tweets, providing a spatial context to mobility patterns.

Sina Weibo, often referred to as the ‘Twitter of China’, has gained popularity as a source of human mobility data, particularly among the Chinese population. With over 300 million active users, Weibo offers valuable insights into various topics, including health-related issues, rumour detection, and urban park evaluation. Zhang and Zhou (2018) conducted a study on recreational visits to urban parks in Beijing using Weibo geotagged data, finding a significant correlation between Weibo check-in data and actual park visitation statistics. Weibo data has also been combined with other datasets to achieve comprehensive research objectives.

Instagram, Flickr, and Tencent User Data (TUD) have also been utilized as sources of human mobility data. Instagram, known for its visual content, has been used to capture user sentiments related to parks and other points of interest. Hu et al. (2014) conducted a study on photo content analysis on Instagram, categorizing photos using computer vision techniques. During the COVID-19 pandemic, Instagram was used to investigate changes in public park visitation, showing a significant decline following lockdown measures.

Flickr, a popular photo-sharing platform, has been used to explore the temporal, spatial, and social dimensions of user behaviour in urban parks. Song et al. (2020) found that photo-user days (PUD) derived from both Flickr and Instagram provide a more accurate representation of visit frequency to urban parks. Researchers can crawl geolocated photos from Flickr using specific keywords generated from park photographs.

Tencent User Data (TUD) has proven to be a robust source of data, especially in China, for various urban studies. TUD provides real-time location data with high spatial resolution, making it suitable for urban planning and resource management. Chen et al. (2018) found that TUD data is effective for measuring urban park usage, given its substantial user base in China. Researchers using TUD can customize the temporal and spatial scope of their data, allowing for flexibility in their analyses. TUD has been employed to understand factors influencing visitation to mini-urban parks in China, revealing insights into how factors like population density and park amenities impact park usage (W. Zhai, Liu, and Peng 2021).

4.2.3. Navigation services

Navigation services, such as Google Maps, Amap, and Baidu Map, have emerged as crucial tools for preserving and disseminating historical locational information, shaping research in various domains (Jiang, Huang, and Li 2021; Shuttleworth and Gould 2023; Sulyok and Walker 2020). In this study, we explore the significance and applications of these services.

The Google Mobility Report, released during the COVID-19 pandemic, offers real-time analytics on human mobility. It derives data from aggregated and anonymized users participating in Google Location History. This data, including mobile phone locations and check-ins, has been extensively utilized in research. It adheres to privacy protocols under the user agreement policy. For example, Geng et al. (2021) used this data to analyse the impact of the pandemic on global urban park visitation, revealing disparities in visitor numbers between economies.

Google Maps, a part of Google’s navigation services, provides users with navigation tools and transportation options. It enables the creation of spatiotemporal movement data based on origin-destination matrix. Chang et al. (2019) employed Google Maps data to study spatial inequality in urban park accessibility in Hong Kong, demonstrating the influence of public transport on park visitation and spatial disparities.

Baidu Map, a popular Chinese navigation service, shares similarities with Google Maps. Despite Google’s wider user coverage, Baidu Map has a significant scholarly presence. It offers features like satellite imagery and route planning, relying on data sources like NavInfo and OpenStreetMap. Researchers have used Baidu Map to map travel flows and investigate social inequalities in park accessibility (W. Liu, Dong, and Chen 2017; Pei et al. 2022b; Qin et al. 2020; M. Xu et al. 2017; Z. Zhou and Xu 2020; Z. Zhou et al. 2023). Its advantage lies in real-time mobility data.

Amap, an online map provider in China, has evolved into a comprehensive travel service platform. It offers route planning, modal choice assistance, and location search services. Researchers have increasingly used Amap’s real-time dynamic data, as an alternative to static data sources. Zhu and Shi (2022) examined job accessibility in Kunshan city, while Niu et al. (2018) assessed the spatial distribution of urban parks in Wuhan. Both studies utilized Amap’s API to obtain travel time and mode data, revealing insights into accessibility and transportation challenges.

4.2.4. Public transit records

Public transit records constitute a valuable form of human mobility data extracted from diverse public transit systems, including trains, buses, ferries, metros, and air flights. According to Hu et al. (2021), public transit data can be classified into two categories: schedule timetable data and actual passenger records. The former provides insights into the operational capacity and timing of each transit mode, facilitating the estimation of ridership (T. Hu et al. 2021). Typically, such data is readily accessible on transit websites or at stations, enabling the public to plan their journeys effectively. On the other hand, actual passenger record data is typically obtained from transit systems when passengers utilize their smart cards to enter the transit. For example, bus transit smart cards capture timestamps denoting trip start and end times, as well as the stations where passengers embark and disembark. Despite the immense potential of public transit systems as sources of human mobility data, their application in urban park accessibility has been severely underexplored. Notably, Li et al. (2021) employed official transit records from Nanjing, China to evaluate park accessibility. This dataset encompassed various information, including traffic speed, estimated time, and route assignment. The authors’ findings indicate that conventional data and methodologies employed for assessing accessibility lack the robustness exhibited by traffic data derived from intelligent public transit systems.

4.3. Methodological and thematic frameworks for evaluating accessibility using human mobility data

In this article, we provide an analysis of the approaches for assessing the accessibility of urban parks using human mobility data, as summarized in Table 1. Our analysis classifies these methodological frameworks into seven overlapping approaches, and they are: 1) inequality and inequity levels, 2) users’ perceptions and exposure levels, 3) frequency and variations in park visitation, 4) park service area (PSA) or effective service radius (ESR), 5) happiness and sentiment levels, 6) travel mode choice or trip assignment, and 7) park characteristics.

Table 1. Methodological and thematic frameworks to evaluate park accessibility using human mobility data.

Studies	Methodological perspective	Modeling tasks	Findings
Xiao et al. (2019) Li et al. (2021) Pie et al. (2022) Xu et al. (2017) Ren and Guan (2022) Guo et al. (2019) Shuqi, Wei and Xinyu (2023)	Inequality and inequity level	Generalized linear model (GLM) including multivariate regression, hierarchical, difference-in-difference model. Spatial autocorrelation, including bivariate correlation, Gini coefficient	Findings show a generalized inequality and inequity in the spatial distribution of parks and park use, including among the elderly population, low-income neighborhoods, and socially disadvantaged minorities.
Yoo and Roberts (2022) Guan et al. (2021) Huang et al. (2022b) Luo et al. (2022) Lyu and Zhang (2019)	Users’ perception or exposure level	Generalized linear models (GLM), including multiple linear regression and linear mixed regression. Spatial statistics, including kernel density estimation (KDE), Density threshold (DT)	Findings indicate a generalized spatial heterogeneous pattern in park use based on users’ perceptual inclinations where park use is influenced by environmental features, park attributes and different available travel modes.
Geng et al. (2021) He et al. (2022) Zhou et al. (2022) Linnell et al. (2022) Song et al. (2022) Rice and Pan (2021) Zhou et al. (2022) Liang and Zhang (2021) Li et al. (2020) Donahue et al. (2018) Suse et al. (2021) Jay et al. (2022)	Frequency and variations in park visitations	Generalized linear models (GLM), including fixed effect regression, multiple linear regression, and stepwise regression. Probabilistic models like Bayesian inference	Findings reveal that strong spatial variation in park visitor numbers exists among different subpopulations and park visitations are influenced by median income, park characteristics, distance, and pandemic and pandemic-related policies.
Guan et al. (2021) He et al. (2022) Chen and Poorthuis (2021) Guo et al. (2019) Zhai et al. (2018) Niu et al. (2018) Guan et al. (2020)	Park service area (PSA) or effective service radius (ESR)	Generalized linear models (GLM), including stepwise regression and multiple linear regressions. Spatial statistics including kernel density estimation (KDE), Density threshold (DT)	Findings generally indicate that the park service area (or effective service radius) varies significantly by case studies (e.g. Beijing between 5 km to 10 km). More importantly, travel distance and time are essential determinants of park accessibility
Schwartz et al. (2022) Kovacs-Györi et al. (2018) Lai and Deal (2023) Schwartz et al. (2022) Song et al. (2020) Song et al. (2020)	Happiness or sentiment level	Natural language processing (NLP) models including Hedonometer, Word2vec, deep learning long short-term memory, language assessment by Mechanical Turk (LabMT)	Findings indicate that park users generally express positive sentiments or happiness levels than non-park users, especially while inside parks. Furthermore, these sentiments are expressed through tweets and photos posted on social media platforms like Instagram, Twitter, and Flickr.
Chang et al. (2019) Liu et al. (2017) Zhou et al. (2022) Zhou and Xu (2020) Tan et al. (2022)	Travel mode choice or trip assignment	Spatial statistics including kriging, autocorrelation, kernel density estimation (KDE)	Findings reveal spatial-temporal travel patterns of how, when and which urban parks are visited by park users by footfalls (or walking), public transport, private cars, and biking.
Qin et al. (2020) Chuang et al. (2022) Li et al. (2020) Volenec et al. (2021) Li et al. (2022) Chen et al. (2018)	Park characteristics	Generalized linear models (GLM), including multiple linear regression. Descriptive and inferential statistics like mean, median, and correlation. Clustering, including principal component analysis	Findings show, among other things, that accessibility is highly influenced by the spatial diversity and characteristics of parks, which determines the spatial diversity and density of check-ins. External factors like population density, park driveway, facilities and infrastructure, and multiple park entrances generally have a positive influence on accessibility.

The perspective of inequality and inequity in relation to urban parks is anchored on the principles of environmental and social justice. This perspective seeks to determine the extent to which socially disadvantaged minority groups, such as the elderly, children, black individuals, and people with disabilities, can access urban parks (Guo, Song, et al. 2019; Shuqi, Wei, and Xinyu 2023). The COVID-19 pandemic further exacerbated the existing inequalities to urban park accessibility due to the widespread implementation of stay-at-home orders. Notably, in the U.S., it was discovered that these demographics and those with lower incomes, faced challenges in travelling longer distances to visit larger parks, as smaller parks were closed to the public as a measure to curb the spread of the virus (Shuqi, Wei, and Xinyu 2023).

An alternative perspective regarding the accessibility of urban parks, as explored within the scholarly discourse, pertains to the perceptions and exposure levels of park users. Researchers frequently leverage ‘big data’ derived from popular social media platforms, such as Twitter (X), Instagram, Tencent QQ, and Flickr, which generate a substantial volume of UGC. Subsequently, this dataset undergoes comprehensive analysis to illuminate insights into the perceptions and experiences of individuals concerning urban parks. In the context of the present investigation, a comprehensive examination was conducted to discern usage patterns of parks, with a particular emphasis on the perceptual inclinations of users and the influence wielded by environmental attributes, park features, and available modes of transportation (Guan et al. 2020; J. H. Huang et al. 2022; Luo et al. 2022; Lyu and Zhang 2019; Yoo and Roberts 2022). The frequency of park visitation and its associated variations were subjected to scrutiny through the meticulous tracking of user visits to parks, thereby furnishing invaluable insights into the utilization of these park spaces. These visitation counts were acquired in real-time through the collection of time-activity data from park-goers located within parks or their proximate areas. The findings derived from this inquiry unveiled several salient facets: firstly, the existence of spatial disparities in visitor numbers among diverse subpopulations, and secondly, the intricate interplay of factors such as median income, park attributes, prevailing pandemic situation, and pandemic-related policies in shaping patterns of park visitation.

Users’ happiness and sentiment levels is another study approach. This methodology involves the utilization of social media platforms and navigation services to assess user expressions, which are transformed into numerical values through various techniques. NLP models within the realm of machine learning are frequently employed to convert textual content into binary representations. The results derived from these investigations consistently indicate that individuals using urban parks tend to exhibit more positive sentiments and higher levels of happiness, especially when they are within the park premises. Popular social media platforms such as Instagram, Twitter, and Flickr provide channels through which users convey their sentiments through posts, including tweets and photographs. Moreover, most of the studies extend users’ sentiment assessment to exploring the modal choice to parks. In these inquiries, data from navigation services, which offer insights into route planning and modal choices, are frequently harnessed (Chang et al. 2019; W. Liu, Dong, and Chen 2017, Tan et al. 2022; Z. Zhou and Xu 2020). The outcomes of these investigations elucidate spatial-temporal travel patterns, shedding light on when, where, and which urban parks are frequented based on the utilization of different modes of transport since users generally express their frustrations or happiness about the distance travelled and modal choice made.

Assessing Park accessibility based on park characteristics involved evaluating the quality and spatial diversity of park amenities, which contribute to the benefits derived from park use. The findings of these studies highlighted the significant influence of park characteristics, spatial diversity, and density of check-ins on accessibility. Factors such as population density, park driveways, facilities, infrastructure, and multiple park entrances were found to positively influence accessibility (Qin et al. 2020; Volenec et al. 2021).

On the modelling tasks employed, generalized linear models (GLMs) such as stepwise regression, difference-in-difference method, multiple linear regression, and multivariate regression were predominantly utilized. Spatial statistics techniques, including kernel density estimation (KDE), spatial autocorrelation, kriging, bivariate correlation, density threshold (DT), and Gini Coefficient Index (GNI), were commonly applied to analyse the data. Furthermore, studies focusing on happiness and sentiment perspectives utilized NLP techniques such as topic modelling (e.g. Hedonometer, Word2vec, deep learning long short-term memory) and LabMT to analyse large-scale textual data. Probabilistic models, including Bayesian inference, were employed to analyse frequency and variations in park visitations. Clustering analysis techniques such as principal component analysis (PCA) was used to reduce the dimensionality of large-scale park characteristics for analysis. Descriptive statistics, including mean and median, were employed to describe users’ demographic data.

5. Discussion

5.1. Empirical findings and implications for park accessibility

The findings offer valuable insights into the accessibility of urban parks and the factors influencing park visitation patterns and wide implications for urban planning, public health, and community well-being. The analysis reveals a notable increase in the utilization of human mobility data for studying urban park accessibility, particularly since 2010. This trend underscores the growing recognition of human mobility as a crucial domain in disciplines such as public health, transportation, and urban studies (Barbosa et al. 2018; Smolak et al. 2021; Zhao et al. 2016). The surge in publications related to urban park accessibility, especially after 2020, highlights the impact of COVID-19 on park visitation behaviours (Rice and Pan 2021; Shuqi, Wei, and Xinyu 2023; Sung et al. 2022). Researchers have extensively investigated the effects of pandemic-related control measures, such as stay-at-home orders, on park visitation dynamics, emphasizing the importance of outdoor spaces for physical and mental well-being during times of crisis (X. Huang et al. 2021, 2022).

By leveraging human mobility data, policymakers and urban planners can identify underserved areas, optimize park siting and design, and tailor interventions to meet the diverse needs of park users (Chuang, Benita, and Tunçer 2022; F. Li et al. 2020; Qin et al. 2020). Moreover, insights into park visitation patterns can inform targeted outreach strategies to promote park usage and facilitate equitable access to green spaces for all members of the community. Overall, the empirical evidence presented in this review underscores the importance of integrating data-driven approaches into park planning and management practices to create inclusive, accessible, and vibrant urban parks.

The geographical distribution of research endeavours reveals a concentration of studies in China, driven by China’s dense urban centres and rich natural landscapes. While China accounts for a significant portion of the reviewed literature, there is relatively less documentation of human mobility data usage for park accessibility in other countries, including the United Kingdom, Japan, Singapore, and South Korea, and mere absence for Africa and South American countries. This geographic disparity underscores the need for a more global perspective on park accessibility research, considering the diverse urban contexts and mobility patterns worldwide.

While social media emerges as the predominant platform for studying park visitation patterns due to its widespread usage and abundant geotagged data, there is a notable absence of transit records in the literature. This underscores the need for further exploration of public transportation’s role in facilitating park access, particularly in urban areas with diverse mobility needs. Additionally, challenges regarding transport data in human mobility datasets, such as the lack of information about transport modes due to anonymization and aggregation, warrant attention.

5.2. Data positioning and representation

The issue of data positioning, particularly in relation to mobile phone data, represents a significant consideration. This issue encompasses two key aspects when examining mobile phone data. Firstly, the limited or poor network signals available to most mobile phone users can result in positional inaccuracies. Factors such as user location, network provider, data package, and special permissions (user consent) can influence the quality of network signals. Consequently, researchers may face challenges in accurately determining the actual user positions during data collection. Mobile base stations typically have a coarse radius ranging from 300 metres to 1 kilometre, further complicating the distinction between park users and non-users. For example, a park user with a weak mobile phone signal or lack of internet connection may not be captured in the data, while a non-park user with a strong network signal and coarse data may be captured due to the mobile network’s proximity to the mobile base station triangulation and park boundary.

Secondly, passive positioning approaches utilize metadata derived from mobile phone users through incoming or outgoing calls or text messages (e.g. call detail records – CDR) that are automatically stored by the network provider. However, the positional accuracy of this metadata varies depending on the country and network, ranging from a few hundred metres to several kilometres. This variability poses challenges in reaching valid conclusions based on the data. Therefore, it is essential to recognize and address the complexities associated with data positioning in human mobility data, particularly mobile phone data, to ensure the reliability and validity of the conclusions drawn from such data.

5.3. Data coverage and stability concerns

Another complexity intertwined with human mobility datasets is the matter of data coverage. No single dataset provides a comprehensive representation of the entire demographic, thereby posing challenges when generalizing about the wider population. For instance, depending on the geographical location, the spatial coverage of users in navigation services can fluctuate between 60% and 90%. In a similar vein, social media platforms do not present uniform coverage rates. This predicament brings forth the query: to which population does the study cater? Despite technological advancements and the prevalent use of social media, there remains a considerable segment of the population that abstains from utilizing social media platforms, navigation services for trips, or even smartphones. Hence, these coverage limitations must be considered when making inferences from human mobility datasets. Also, not forgetting technological disparities can affect the timetable of data availability between developed and developing countries.

The study identified data stability as a significant issue. Across all 45 papers reviewed, substantial temporal variations were evident in data related to various mobility types, presenting a considerable challenge when attempting to compare results. An associated concern is the inconsistency in open data policies, which fluctuates due to factors such as ownership transfers (as observed in the case of Twitter becoming ‘X’), government sanctions (including outright or temporary bans), and court injunctions (as seen in Twitter litigations). These fluctuations prompt researchers to question the reliability of big data. How resilient is this data in the face of adversity, and can the data platform be trusted consistently over an extended period?

5.4. Reproducibility and repeatability

The analytical framework employed to assess human mobility data plays a crucial role in obtaining precise spatiotemporal insights into people’s movement patterns. However, ensuring reproducibility and repeatability of these analyses is imperative. Chen and Poorthuis (2021) rightly observed in their study that many publications on human mobility data lack detailed discussions and disclosures of the source codes of the algorithms used to analyse this vast amount of data. This omission makes it challenging to compare algorithms and replicate results. Furthermore, the reviewed literature for this project revealed a lack of disclosure of Python scripts and algorithms utilized for scraping various APIs, further hindering the replication of studies and the attainment of similar outcomes.

The absence of standardized practices and ethical guidelines for analysing big human mobility data contributes to the perception that the analytical framework heavily relies on the expertise and knowledge of individual researchers rather than established conventions. Consequently, the reproducibility and repeatability of analyses are compromised. Additionally, it was observed in some studies that the spatial grids employed to scale the study area using mobile phone data varied significantly, ranging from 300 metres to 500 metres (Niu et al. 2018).

5.5. Ethical issues around human mobility data use in urban park accessibility

As we have observed the significant utility of human mobility data in park accessibility research, it is essential to acknowledge the emergence of various ethical concerns associated with their increasing generation and utilization. Among these concerns that we have identified, key ones include data privacy and protection, data consent, and data ownership.

Privacy has become the paramount concern in the era of extensive big data utilization. The sheer volume of data collected and analysed often poses a threat to individual data privacy. Given the vastness of big data, many researchers, who may not be proficient in data curation or lack in-depth programming skills, often rely on external sources like GitHub for code and debugging. Unfortunately, this practice can inadvertently expose sensitive user information to potential attacks. Although efforts have been made to promote data privacy, such as the establishment of the International Data Responsibility Group and the implementation of data protection laws in the US and EU, concerns persist, particularly in developing countries (Taylor 2016).

Data consent is closely intertwined with data privacy. While individuals may consent to ‘user agreement policies’ when using social media and other data-generating platforms, they often remain unaware of how their private information is actually used. In many cases, users do not explicitly grant permission for their data to be used by third parties, such as research institutions or individuals, yet their data is acquired without direct consent. Moreover, numerous studies involve the direct scraping of user data through web crawling from platforms like Baidu and Amap, raising substantial ethical concerns.

Data ownership represents another complex issue that demands attention. A contentious debate revolves around the rightful owners of the data, whether it is the users themselves or the data curators, and who should wield control over monetization. Numerous concerns have arisen regarding data curators profiting from users’ data without adequately informing the individuals from whom this data is collected. For instance, a highly contentious practice is exemplified by SafeGraph, which monetizes data gathered from mobile phone users without full transparency, a practice often deemed unethical.

5.6. Outlook for the adoption of human mobility data in park design

The field of utilizing human mobility data for park accessibility design holds immense promise for enhancing park spaces and fostering community well-being. Building upon the insights from the reviewed literature, the outlook for the future presents potential avenues for research, practice, and policy. The role of data analytics techniques, such as machine learning and spatiotemporal analysis, in understanding park visitation patterns and informing design decisions is still understudied. Future research endeavours should focus on integrating cutting-edge analytics methodologies to extract actionable insights from large-scale human mobility datasets. By harnessing the predictive power of machine learning models and the spatial-temporal dynamics captured in the data, planners, and policymakers can optimize park layouts, amenities, and programming to better meet the evolving needs of urban populations.

Secondly, addressing disparities in park accessibility and usage remains a critical challenge for urban planners and policymakers. Future efforts should prioritize equity and inclusivity in park design, leveraging human mobility data to identify underserved communities and implement targeted interventions as it is in transport planning for recommending autonomous vehicles, public transit, and rider-share. By adopting a data-driven approach to equity planning, cities can ensure that all residents have equitable access to high-quality green spaces and recreational opportunities, regardless of socio-demographic factors.

6. Conclusion

The exploration of urban park accessibility, utilizing the vast potential of human mobility data, stands as a critical area of research, addressing the shortcomings of conventional static data sources such as census data. The expanding scope and inherent complexity of human mobility trajectories necessitate the development of advanced methodologies to enable real-time spatiotemporal analysis. In our thorough review, we investigate a wide array of human mobility data types utilized in assessing park accessibility. Our examination spans a diverse range of data sources, including mobile phone data, social media platforms, navigation services, and public transit systems. We have methodically organized the methodological framework into seven distinct categories, thereby delineating the methodological evolution in the application of human mobility data for accessibility analysis. While the use of human mobility data offers significant promise, we also acknowledge and address the ongoing challenges associated with this approach. These include concerns over data privacy, limitations in data coverage, issues related to the positioning and representation of data, and the imperative for reproducibility and repeatability in research endeavours. We place a strong emphasis on the ethical aspects of big data utilization and advocate for responsible data management practices to address these challenges effectively. Our review acts as a comprehensive guide, shedding light on the current state of research while pinpointing opportunities for future inquiry. We highlight the necessity for innovative methods that transcend traditional data approaches, leveraging the capabilities of human mobility data to create a more dynamic, precise, and ethical framework for analysing urban park accessibility. This progressive approach has the potential to not only advance the field of urban studies but also to guide more adaptive and sustainable urban planning and policymaking.

Disclosure statement

No potential conflict of interest was reported by the author(s).